Download CDM article on quantum chaos - Department of Mathematics

RECENT DEVELOPMENTS IN MATHEMATICAL QUANTUM CHAOS
PRELIMINARY VERSION/COMMENTS APPRECIATED
STEVE ZELDITCH
Abstract. This is a survey of recent results on quantum ergodicity, specifically on the large
energy limits of matrix elements relative to eigenfunctions of the Laplacian. It is mainly
devoted to QUE (quantum unique ergodicity) results, i.e. results on the possible existence
of a sparse subsequence of eigenfunctions with anomalous concentration. We cover the lower
bounds on entropies of quantum limit measures due to Anantharaman, Nonnenmacher, and
Rivière on compact Riemannian manifolds with Anosov flow. These lower bounds give new
constraints on the possible quantum limits. We also cover the non-QUE result of Hassell in
the case of the Bunimovich stadium. We include some discussion of Hecke eigenfunctions
and recent results of Soundararajan completing Lindenstrauss’ QUE result, in the context of
matrix elements for Fourier integral operators. Finally, in answer to the potential question
‘why study matrix elements’ it presents an application of the author to the geometry of
nodal sets.
Contents
1. Wave equation and geodesic flow
2. Weyl law and local Weyl law
3. Invariant states defined by eigenfuntions and their quantum limits
4. Quantum ergodicity and mixing of eigenfunctions
5. Concentration of eigenfunctions around hyperbolic closed geodesics
6. Boundary quantum ergodicity and quantum ergodic restriction
7. Hassell’s scarring result for stadia
8. Matrix elements of Fourier integral operators
9. QUE of Hecke eigenfunctions
10. Variance estimates: Rate of quantum ergodicity and mixing
11. Entropy of quantum limits on manifolds with Anosov geodesic flow
12. Applications to nodal hypersurfaces of eigenfunctions
References
4
15
17
23
28
32
35
39
41
45
48
66
71
Quantum chaos on Riemannian manifolds (M, g) is concerned with the asymptotics of
eigenvalues and orthonormal bases of eigenfunctions
(1)
∆ϕj = λ2j ϕj ,
hϕj , ϕk i = δjk
of the Laplacian ∆ when the geodesic flow g t : Sg∗ M → Sg∗ M is ergodic or Anosov or in some
other sense chaotic. Model examples include compact or finite volume hyperbolic surfaces
Date: November 22, 2009.
Research partially supported by NSF grant DMS 0904252.
1
2
STEVE ZELDITCH
XΓ = Γ\H, where H is the hyperbolic plane and Γ ⊂ P SL(2, R) is a discrete subgroup. Of
special interest are the arithmetic quotients when Γ is a discrete arithmetic subgroup. Other
model example include Euclidean domains with ergodic billiards such as the Bunimovich
stadium. The general question is, how does the dynamics of the geodesic (or billiard) flow
make itself felt in the eigenvalues and eigenfunctions (or more general solutions of the wave
equation), which by the correspondence principle of quantum mechanics must reproduce the
classical limit as the eigenvalue tends to infinity?
To avoid confusion, we emphasize that we denote eigenvalues of the Laplacian by λ2j . They
are usually viewed as energies Ej = λ2j . Their square roots λj are called the frequencies.
The most basic quantities testing the asymptotics of eigenfunctions are the matrix elements
hAϕj , ϕj i of pseudo-differential operators relative to the orthonormal basis of eigenfunctions.
As explained below, these matrix elements measure the expected value of the observable A in
the energy state ϕj . Much of the work in quantum ergodicity since the pioneering work of A.
I. Schnirelman [Sh.1] is devoted to the study of the limits of hAϕj , ϕj i as λj → ∞. Although
difficult to determine, these limits are still the most accessible aspects of Reigenfunctions. A
sequence of eigenfunctions is called ergodic or diffuse if the limit tends to S ∗ M σA dµL where
g
σA is the prinicipal symbol of A and dµL is the Liouville measure.
The first priority of this survey is to cover the important new results on the so-called
scarring or QUE (quantum unique ergodicity) problem. Roughly speaking, the problem is
whether every orthonormal basis becomes equidistributed in phase space with respect to
Liouville measure as the eigenvalue tends to infinity, i.e.
Z
hAϕj , ϕj i →
σA dµL
Sg∗ M
or whether there exists a sparse exceptional sequence which has a singular concentration. One
series of positive results due to N. Anantharaman [A], N. Anantharaman-S. Nonnenmacher
[AN] (see also [ANK]), and G. Rivière [Riv] flows from the study of entropies of quantum
limit measures and are based on a difficult and complex microlocal PDE analysis of the long
time behavior of the wave group on manifolds with Anosov geodesic flow. These results given
lower bounds on entropies of quantum limit measures which imply that the limit measures
must be to some extent diffuse. Another series of results due R. Soundararajan [Sound1]
use L-function methods to complete the near QUE result of E. Lindenstrauss [LIND] for
Hecke eigenfunctions in the arithmetic case. In the negative direction, A. Hassell [Has] uses
microlocal methods to proved the long-standing conjecture that generic stadia are not QUE
but rather have exceptional sequences of bouncing ball type modes which concentrate or scar
along a Lagrangian submanifold.
QUE is not the only interesting problem in quantum chaos. Another series of results in the
arithmetic case are the remarkably sharp Luo-Sarnak asymptotics [LS, LS2] of the variances
Z
X
|hAϕj , ϕj i −
σA dµL |2
j:λj ≤λ
Sg∗ M
for holomorphic Hecke eigenforms and their recent generalization to smooth Maass forms by
Zhao [Zh]. Physicists (see e.g. [FP]) have speculated that the variance is related to the classical auto-correlation function of the geodesic flow. The Luo-Sarnak-Zhao results partially
RECENT DEVELOPMENTS IN MATHEMATICAL QUANTUM CHAOS
3
confirm this conjecture in the special case of arithmetic surfaces and Hecke eigenfunctions,
but correct it with an extra arithmetic factor.
A final direction we survey is the applications of quantum ergodicity to problems in nodal
geometry. Given the large amount of work on matrix elements hAϕj , ϕj i, it is natural for
geometric or PDE analysts to ask how one can use such matrix elements to study classical
problems on eigenfunctions such as growth and distributions of zeros and critical points.
One of our themes is that for real analytic (M, g), logarithms log |ϕCj (ζ)|2 of analytic continuations of ergodic eigenfunctions are asymptotically ‘maximal pluri-subharmonic functions’
in Grauert tubes and as a result their zero sets have a special limit distribution [Z5]. Thus,
ergodicity causes maximal oscillation in the real and complex domain and gives rise to special
distributions of zeros and (conjecturally) to critical points.
In preparing this survey, we were only too aware of the large number of surveys and
expository articles that already exist on the QUE and entropy problems, particularly [AN2,
ANK, CV3, LIND2, Sar3] (and for earlier work [Z1, Z2, M, Sar]). To a large degree we closely
follow the original sources, emphasizing intuition over the finer technical details. Although
we go over much of the same material in what are now standard ways, we also go over
some topics that do not seem as well known, and possibly could be improved, and we fill in
some gaps in the literature. In particular, most expositions emphasize the quantum entropic
uncertainty principle of [AN] due to its structural nature. So instead we emphasize the ideas
of [A], which are less structural but in some ways more geometric.
One of the novelties is the study in §8 of matrix elements hF ϕj , ϕj i of Fourier integral
operators F relative to the eigenfunctions. Hecke operators are examples of such F but
restrictions to hypersurfaces provide a different kind of F which have recently come up
in quantum ergodic restriction theory [TZ3]. We give a rather simple result on quantum
limits for Fourier integral operators which answers the question, “what invariance property
do quantum limit measures of Hecke eigenfunctions possess?”. Hecke analysts have worked
with a partial quasi-invariance principle due to Rudnick-Sarnak, and although the exact
invariance principle may not simplify the known proofs of QUE it is of interest to know that
one exists. We plan to put complete proofs in a future article.
Another aspect of eigenfunctions is their possible concentration of eigenfunctions around
closed hyperbolic geodesic. Mass concentration within thin or shrinking tubes around such
geodesics has been studied by S.Nonnenmacher-A.Voros [NV], Y. Colin de Verdière-B. Parisse
[CVP], J. Toth and the author [TZ2] and outside of such tubes by N. Burq-M. Zworski [BZ]
and H. Christianson [Chr]. Very striking and surprising studies of analogous behavior in the
‘quantum cat map’ setting are in [FNB, FN].
A few words on the limitations of this survey. Due to the author’s lack of knowledge, the
L-function or arithmetic methods in QUE will not be discussed in more than a cursory way.
Rather we concentrate on the PDE or microlocal aspects of quantum ergodicity. By necessity,
microlocal analytical proofs must make direct connections between spectrum and geodesic
flow, and cannot by-pass this obstacle by taking a special arithmetic route. Moreover, we
restrict the setting to Laplacians on Riemannian manifolds for simplicity of exposition, but
it is just a special case of the semi-classical asymptotics (as the Planck constant tends to
zero) of spectra of Schrödinger operators h2 ∆ + V . Schrödinger eigenfunctions are equally
relevant to physicists and mathematics. Another important setting is that of quantum
‘cat maps’ on tori and other Kähler manifolds. In this context, the quantum dynamics
4
STEVE ZELDITCH
is defined by quantizations of symplectic maps of Kähler manifolds acting on spaces of
holomorphic sections of powers of a positive line bundle. The mathematics is quite similar
to the Riemannian setting, and results in the cat map setting often suggest analogues in
the Riemannian setting. In particular, the scarring results of Faure-Nonnemacher- de Bièvre
[FNB] and the variance results of Kurlberg-Rudnick [KR] and Schubert [Schu3] are very
relevant to the scarring and variance results we present in the Riemannian setting.
Although we do not discuss it here, a good portion of the literature of quantum chaos is
devoted to numerics and physicists’ heuristics. Some classics are [Ber, He, He2]. The article
of A. Barnett [Bar] (and in the cat map case [FNB, FN] ) is a relatively recent discussion of
the numerical results in quantum ergodicity. There is a wealth of phenomenology standing
behind the rigorous results and heuristic proofs in this field. Much of it still lies far beyond
the scope of the current mathematical techniques.
Finally, we thank N. Anantharaman, A. Hassell, and S. Nonnenmacher for explanations
of important points in their work. We have incorporated many of their clarifications in the
survey. We also thank K. Burns, H. Hezari, J. Franks, E. Lindenstrauss, P. Sarnak, A.
Wilkinson and J. Wunsch for further clarifications and corrections. Of course, responsibility
for any errors that remain is the author’s.
1. Wave equation and geodesic flow
Classical mechanics is the study of Hamiltonians (real valued functions) on phase space
(a symplectic manifold) and their Hamiltonian flows. Quantum mechanics is the study of
Hermitian operators on Hilbert spaces and the unitary groups they generate. The model
quantum Hamiltonians we will discuss are Laplacians ∆ on compact Riemannian manifolds
(M, g) (with or without boundary). Throughout we denote the dimension of M by d =
dim M .
1.1. Geodesic flow and classical mechanics. The classical phase space in this
P setting is
the cotangent bundle T ∗ M of M , equipped with its canonical
symplectic
form
i dxi ∧ dξi .
qP
n
∗
ij
The metric defines the Hamiltonian H(x, ξ) = |ξ|g =
ij=1 g (x)ξi ξj on T M , where
gij = g( ∂x∂ i , ∂x∂ j ), [g ij ] is the inverse matrix to [gij ]. We denote the volume density of (M, g)
by dV ol and the corresponding inner product on L2 (M ) by hf, gi. The unit (co-) ball bundle
is denoted B ∗ M = {(x, ξ) : |ξ| ≤ 1}.
The classical evolution (dynamics) is given by the geodesic flow of (M, g), i.e. the Hamiltonian flow g t of H on T ∗ M . By definition, g t (x, ξ) = (xt , ξt ), where (xt , ξt ) is the terminal
tangent vector at time t of the unit speed geodesic starting at x in the direction ξ. Here and
below, we often identify T ∗ M with the tangent bundle T M using the metric to simplify the
geometric description. The geodesic flow preserves the energy surfaces {H = E} which are
the co-sphere bundles SE∗ M . Due to the homogeneity of H, the flow on any energy surface
{H = E} is equivalent to that on the co-sphere bundle S ∗ M = {H = 1}.
We define the Liouville measure µL on S ∗ M to be the surface measure dµL = dxdξ
induced
dH
by the Hamiltonian H = |ξ|g and by the symplectic volume measure dxdξ on T ∗ M . The
geodesic flow (like any Hamiltonian flow) preserves dxdξ. It also preserves Liouville measue
on S ∗ M .
RECENT DEVELOPMENTS IN MATHEMATICAL QUANTUM CHAOS
5
1.2. Ergodic and Anosov geodesic flows. The geodesic flow is called ergodic if the
unitary operator
(2)
V t a(ρ) = a(g t (ρ))
on L2 (S ∗ M, dµL ) has no invariant L2 functions besides constant functions. Equivalently, any
invariant set E ⊂ S ∗ M has either zero measure or full measure.
A geodesic flow g t is called Anosov on on Sg∗ M if the tangent bundle T Sg∗ M splits into g t
invariant sub-bundles E u (ρ) ⊕ E s (ρ) ⊕ R XH (ρ) where E u is the unstable subspace and E s
the stable subspace. They are defined by
||dg t v|| ≤ Ce−λt ||v||, ∀v ∈ E s , t ≥ 0,
||dg t v|| ≤ Ceλt ||v||,
∀v ∈ E u , t ≤ 0.
The sub-bundles are integrable and give stable, resp. unstable foliations W s , W u . The
leaves through x are denoted by W s (x), W u (x). Thus, the geodesic flow contracts everything
exponentially fast along the stable leaves and expands everything exponentially fast along
the unstable leaves. Anosov geodesic flows are ergodic. We refer to [Kl] for background.
The unstable Jacobian J u (ρ) at ρ is defined by
¡ −1
¢
(3)
J u (ρ) = det dg|E
u (g 1 ρ) .
Define
(4)
Λ = − supρ∈S ∗ M log ||J u (ρ)||
¯R
¯
= inf µ∈MI ¯ S ∗ M log J u (ρ)dµ(ρ)¯ > 0.
If (M, g) is of dimension d and constant curvature −1 then Λ = d − 1. Here, MI is the set
of invariant probability measures for g t .
The maximal expansion rate of the geodesic flow is defined by
1
(5)
λmax = lim log sup ||dgρt ||.
t→∞ t
ρ∈S ∗ M
The derivative dg t may be expressed in terms of Jacobi fields (see [Kl], Lemma 3.1.17).
In the hyperbolic (Anosov case), the stable/unstable sub-bundles are spanned by unstable
u
Jacobi fields Y1u , . . . , Yd−1
whose norms grow exponentially as t → +∞ and decay exponens
tially as t → −∞, resp. stable Jacobi fields Y1s , . . . , Yd−1
whose norms behave in the opposite
u
u
fashion. The unstable Jacobian is the norm ||Y1 ∧ · · · ∧ Yd−1
||.
1.3. Eigenfunctions
and eigenvalues of ∆. The quantization of the Hamiltonian H is
√
the square root ∆ of the positive Laplacian,
n
1 X ∂ ij ∂
∆ = −√
g g
g i,j=1 ∂xi
∂xj
of (M, g). Here, g = det[gij ]. The eigenvalue problem on a compact Riemannian manifold
∆ϕj = λ2j ϕj ,
hϕj , ϕk i = δjk
is dual under the Fourier transform to the wave equation. Here, {ϕj } is a choice of orthonormal basis of eigenfunctions, which is not unique if the eigenvalues have multiplicities > 1.
6
STEVE ZELDITCH
The individual eigenfunctions are difficult to study directly, and so one generally forms the
spectral projections kernel,
X
(6)
E(λ, x, y) =
ϕj (x)ϕj (y).
j:λj ≤λ
Semi-classical asymptotics is the study of the λ → ∞ limit of the spectral data {ϕj , λj } or
of E(λ, x, y).
1.4. Eigenvalues and Planck’s constant. Although it is only a notational issue in this
setting, it is conceptually useful to identify inverse frequencies with Planck’s constant, i.e.
to define
(7)
~j = λ−1
j .
The eigenvalue problem then takes the form of a Schrödinger equation
~2j ∆ϕj = ϕj .
Thus the high frequency limit is a special case of the semi-classical limit ~ → 0.
1.5. Eigenvalue Multiplicities. The multiplicity of an eigenvalue λ2j is the dimension
(8)
m(λj ) = dim ker(∆ − λ2j )
of the eigenspace. As will be discussed in §2, the general Weyl law gives the bound m(λj ) ≤
Cλd−1
in dimension d. In the case of negatively curved manifolds, the bound has been
j
improved by a (log λj )−1 factor. Thus it is possible for sequences of eigenvalues to have
the huge multiplicities λd−1
(log λj )−1 . When such multiplicities occur, it is possible to build
j
up superpositions of eigenfunctions which have special concentration properties (see §5).
Hassell’s scarring theorem is based on the non-existence of such multiplicities for certain
sequences of eigenvalues (§7).
In fact such multiplicities do occur in the quantum cap map setting for sparse sequences
of eigenvalues and are responsible for the very strange behavior of eigenstates in that case
[FNB, FN]. It is unknown if such sequences exist in the Riemannian Anosov setting. This
is one of the fundamental obstacles to understanding whether or how scarring occurs in the
Riemannian case.
1.6. Quantum evolution. Quantum evolution is given by the wave group
√
U t = eit
∆
,
which in a rigorous sense is the √quantization of the geodesic flow Φt . It is generated by
the pseudo-differential operator ∆, as defined by the spectral theorem (it has the same
eigenfunctions as ∆ and the eigenvalues λj ). The (Schwartz) kernel of the wave group on a
compact Riemannian manifold can be represented in terms of the spectral data by
X
(9)
U t (x, y) =
eitλj ϕj (x)ϕj (y),
R
j
or equivalently as the Fourier transform R eitλ dE(λ, x, y) of the spectral projections. Hence
spectral asymptotics is often studied through the large time behavior of the wave group.
RECENT DEVELOPMENTS IN MATHEMATICAL QUANTUM CHAOS
7
Evolution can be studied on the level of ‘points’ or on the level of observables. Evolution of
points in classical mechanics gives the orbits or trajectories of the geodesic flow (i.e. the parameterized geodesics). Evolution ψt = U t ψ of wave functions gives the ‘Schrödinger picture’
of quantum mechanics. Eigenfunctions arise in quantum mechanics as stationary states, i.e.
states ψ for which the probability measure |ψ(t, x)|2 dvol is constant where ψ(t, x) = U t ψ(x)
is the evolving state. This follows from the fact that
U t ϕk = eitλk ϕk
(10)
and that |eitλk | = 1. By unitarity U t∗ = U −t .
As an alternative one could define the quantum evolution as in [AN] to be the semi-classical
Schrödinger propagator
U~t := e−it~∆ .
(11)
This replaces the homogeneous Fourier integral operator U t by a semi-classical Fourier integral operator. The theories are quite parallel; we mainly keep to the homogeneous theory in
this survey (See [DSj] and [EZ]) for background on these notions).
1.7. Observables. In the classical setting, observables are (real-valued) functions on T ∗ M
which we usually take to be homogeneous of degree zero (or as zeroth order symbols). In
the quantum setting, observables are ψDO’s (pseudodifferential operators) of all orders; we
often restrict to the subalgebra Ψ0 of ψDO’s of order zero. We denote by Ψm (M ) the
subspace of pseudodifferential operators of order m. The algebra is defined by constructing
a quantization Op from an algebra of symbols a ∈ S m (T ∗ M ) of order m (polyhomogeneous
functions on T ∗ M \0) to Ψm . A function on T ∗ M \0 is called polyhomogeneous or a classical
m
symbol of order m, a ∈ Sphg
, if it admits an asymptotic expansion,
a(x, ξ) ∼
∞
X
am−j (x, ξ), where ak (x, τ ξ) = τ k ak (x, ξ), (|ξ| ≥ 1)
j=0
for some m (called its order). The asympotics hold in the sense that
a−
R
X
am−j ∈ S m−R−1 ,
j=0
k
j−k
|Dxα Dξβ σ(x, ξ)|
where σ ∈ S if sup(1 + |ξ|)
< +∞ for all compact set and for all α, β, j.
There is a semi-classical analogue in the ~ setting where the complete symbol a~ (x, ξ) depends
on ~ in a similar poly-homogeneous way,
a~ ∼ ~
−m
R
X
~j am−j .
j=0
We refer to [DSj, GSt, EZ, Dui] for background on microlocal analysis (or semi-classical
analysis), i.e. for the theory of pseudo-differential and Fourier integral operators.
The main idea is that such observables have good classical limits. The high frequency
limit is mirrored in the behavior of the symbol as |ξ| → ∞.
An useful type of observable is a smooth cutoff χU (x, ξ) to some open set U ⊂ S ∗ M (it
is homogeneous so it cuts off to the cone through U in T ∗ M ). That is, a function which is
one in a large ball in U and zero outside U . Then Op(χU ) is called a microlocal cutoff to
8
STEVE ZELDITCH
U . Its expectation values in the states ϕk give the phase space mass of the state in U or
equivalently the probability amplitude that the particle is in U . (The modulus square of the
amplitude is the probability).
There are several standard useful notations for quantizations of symbols a to operators A
that we use in this survey. We use the notation Op(a) or â interchangeably.
1.8. Heisenberg uncertainty prinicple. The Heisenberg uncertainty principle is the
heuristic principle that one cannot measure things in regions of phase space where the product
of the widths in configuration and momentum directions is ≤ ~.
It is useful to microlocalize to sets which shrink as ~ → 0 using ~ (or λ)-dependent
cutoffs such as χU (λδ (x, ξ)). The Heisenberg uncertainty principle is manifested in pseudodifferential calculus in the difficulty (or impossibility) of defining pseudo-differential cut-off
operators Op(χU (λδ (x, ξ)) when δ ≥ 12 . That is, such small scale cutoffs do not obey the usual
rules of semi-classical analysis (behavior of symbols under composition). The uncertainty
principle allows one to study eigenfunctions by microlocal methods in configuration space
balls B(x0 , λ−1
j ) of radius ~ (since there is no constraint on the height) or in a phase space
−1/2
ball of radius λj .
1.9. Matrix elements and Wigner distributions. These lectures are mainly concerned
with the matrix elements hOp(a)ϕj , ϕk i of an observable relative to the eigenfunctions. The
diagonal matrix elements
(12)
ρj (Op(a)) := hOp(a)ϕj , ϕj i
are interpreted in quantum mechanics as the expected value of the observable Op(a) in the
energy state ϕj . The off-diagonal matrix elements are interpreted as transition amplitudes.
Here, and below, an amplitude is a complex number whose modulus square is a probability.
If we fix the quantization a → Op(a), then the matrix elements can be represented by
Wigner distributions. In the diagonal case, we define Wk ∈ D0 (S ∗ M ) by
Z
(13)
adWk := hOp(a)ϕk , ϕk i.
S∗M
1.10. Evolution of Observables. Evolution of observables is particularly relevant in quantum chaos and is known in physics as the ‘Heisenberg picture’.
The evolution of observables in the Heisenberg picture is defined by
(14)
αt (A) := U t AU −t ,
A ∈ Ψm (M ).
and since Dirac’s Principles of Quantum Mechanics, it was known to correspond to the
classical evolution
(15)
V t (a) := a ◦ g t
of observables a ∈ C ∞ (S ∗ M ). Egorov’s theorem is the rigorous version of this correspondence: it says that αt defines an order-preserving automorphism of Ψ∗ (M ), i.e. αt (A) ∈
Ψm (M ) if A ∈ Ψm (M ), and that
(16)
σUt AUt∗ (x, ξ) = σA (g t (x, ξ)) := V t (σA ), (x, ξ) ∈ T ∗ M \0.
This formula is almost universally taken to be the definition of quantization of a flow or map
in the physics literature.
RECENT DEVELOPMENTS IN MATHEMATICAL QUANTUM CHAOS
9
1.11. Hadamard parametrix. The link between spectral theory and geometry, and the
source of Egorov’s theorem for the wave group, is the construction of a parametrix (or WKB
formula) for the wave kernel. For small times t, the simplest is the Hadamard parametrix,
Z ∞
∞
X
d−3
2
2
t
(t < inj(M, g))
(17)
U (x, y) ∼
eiθ(r (x,y)−t )
Uk (x, y)θ 2 −k dθ
0
k=0
1
where r(x, y) is the distance between points, U0 (x, y) = Θ− 2 (x, y) is the volume 1/2-density,
inj(M, g) is the injectivity radius, and the higher Hadamard coefficients are obtained by
solving transport equations along geodesics. The parametrix is asymptotic to the wave
kernel in the sense of smoothness, i.e. the difference of the two sides of (17) is smooth. The
relation (17) may be iterated using (U t )m = U tm to obtain a parametrix for long times. For
manifolds without conjugate points, the Hadamard parametrix is globally well defined on
the universal cover and can then be summed over the deck transformation group to obtain
a parametrix on the quotient.
Egorov’s theorem (16) may be proved by explicitly writing out the integral representation
for U −t AU t using (17) and by applying the stationary phase method to simplify the integral.
1.12. Ehrenfest time. The aim in quantum chaos is to obtain information about the high
energy asymptotics as λj → ∞ of eigenvalues and eigenfunctions by connecting information
about U t and g t . The connection comes from comparing (9) and (17), or by using Egorov’s
theorem (16). But to use the hypothesis that g t is ergodic or chaotic, one needs to exploit the
connection as ~ → 0 and t → ∞. The difficulty in quantum chaos is that the approximation
of U t by g t is only a good one for t less than the Eherenfest time
(18)
TE =
log |~|
,
λmax
where λmax is defined in (5).
Roughly speaking, the idea is that the evolution of a well constructed “coherent” quantum
state or particle is a moving lump that “tracks along” the trajectory of a classical particle up
to time TH and then slowly falls apart and stops acting like a classical particle. Numerical
studies of long time dynamics of wave packets are given in works of E. J. Heller [He, He2]
and rigorous treatments are in Bouzouina-Robert [BouR], Combescure-Robert [CR] and
Schurbert [Schu3].
The basic result expressed in semi-classical notation is that there exists Γ > 0 such that
(19)
||U~−t Op(a)U~t − Op(a ◦ g t )|| ≤ C~ etΓ .
Such an estimate has long been known if Γ is not specified in terms of the geodesics. It is
implicit in [Be] and used explicitly in [Z4], based on long time wave equation estimates of
Volovoy. The best constant should be Γ = λmax .
Thus, one only expects good joint asymptotics as ~ → 0, T → ∞ for t ≤ TE . As a result,
one can only exploit the approximation of U t by g t for the relatively short time TE .
1.13. Modes, quasi-modes and coherent states. This survey is mainly concerned with
eigenfunctions. But one of the main tools for studying eigenfunctions is the construction
of quasi-modes or approximate eigenfunctions. Lagrangian and coherent states are closely
related to quasi-modes. We refer to [Dui, CV2, B.B, Ra, Ra2] for background.
10
STEVE ZELDITCH
A quasi-mode {ψk } of order zero is a sequence of L2 -normalized functions satisfying
(20)
||(∆ − µk )ψk ||L2 = O(1),
for a sequence of quasi-eigenvalues µk . By the spectral theorem, there must exist true
eigenvalues in the interval [µk − δ, µk + δ] for some δ > 0. Moreover, if Ek,δ denotes the
spectral projection for the Laplacian corresponding to this interval, then
||Ek,δ ψk − ψk ||L2 = O(k −1 ).
A more general definition is that a quasi-mode of order s is a sequence satisfying
(21)
||(∆ − µk )ψk ||L2 = O(µ−s
k ).
Then
||Ek,δ ψk − ψk ||L2 = O(k −s−1 ).
This definition allows for very weak versions of quasi-modes, e.g. quasi-modes of order −1.
They carry no information about eigenvalues since there always exist eigenvalues in intervals
of width one. However, they do carry some information on eigenfunctions.
In references such as [Dui], quasi-modes are constructed in the form of oscillatory integrals
or Lagrangian states,
Z
ψλ (x) =
eiλS(x,ξ) a(x, ξ)dξ.
Rk
The phase generates the Lagrangian submanifold
ΛS = {(x, dx S) : dξ S(x, ξ) = 0} ⊂ T ∗ M.
If Λ ⊂ S ∗ M is a closed Lagrangian submanifold invariant under the geodesic flow then ψλ
is a quasi-mode of order −1. In favorable circumstances, one can further determine the
amplitude so that ||(−∆ − λ2 )ψλ || = O(λ−s ) and construct a high order quasi-mode. Such
a quasi-mode is the ‘quantization’ of Λ and the associated sequence concentrates on Λ.
A coherent state is a special kind of Lagrangian state, or more precisely isotropic state,
which quantizes a ‘point’ in phase space rather than a Lagrangian submanifold. The simi
|x−x0 |2
on Rn , which are localized
plest examples are the states ϕx~0 ,ξ0 (x) = ~−n/2 e ~ hx,ξ0 i e− ~
to the smallest possible volume in phase space, namely a unit cube of volume ~ around
(x0 , ξ0 ). They are Lagrangian states with complex phases associated to positive Lagrangian
submanifolds of complexified phase space.
1.14. Riemannian random waves. Eigenfunctions in the chaotic case are never similar
to such quasi-modes or Lagrangian states. Rather, they are conjectured to resemble random
waves. This heuristic was first proposed by M. V. Berry [Ber], who had in mind random
Euclidean plane waves. The specific model discussed here was used by the author to show
that eigenfunctions in the ergodic case behave to some degree like random orthonormal bases
of spherical harmonics. We use the term Riemannian random waves for this model.
We decompose the spectrum of a compact Riemannian manifold (M, g) into intervals
IN = [N, N + 1] and define the finite dimensional Hilbert spaces HIN as the space of linear
combinations of eigenfunctions with frequencies in an interval IN := [N, N + 1]. The precise
decomposition of R into intervals is not canonical and any choice of intervals of finite width
would work as long as the set of closed geodesics of (M, g) has measure zero. We will not
N
discuss the other case since it has no bearing on quantum ergodicity. We denote by {ϕN j }dj=1
RECENT DEVELOPMENTS IN MATHEMATICAL QUANTUM CHAOS
11
an orthonormal basis of HIN where dN = dim HIN . We endow the real vector space HIN
with the Gaussian probability measure γN defined by
µ ¶dN /2
dN
X
dN
2
(22)
γN (s) =
e−dN |c| dc ,
ψ=
cj ϕN j , dN = dim HIN .
π
j=1
Here, dc is dN -dimensional real Lebesgue measure. The normalization is chosen so that
EγN hψ, ψi = 1, where EγN is the expected value with respect to γN . Equivalently, the dN
real variables cj (j = 1, . . . , dN ) are independent identically distributed (i.i.d.) random
variables with mean 0 and variance 2d1N ; i.e.,
1
δjk .
dN
We note that the Gaussian ensemble is equivalent to picking ψN ∈ HIN at random from the
unit sphere in HIN with respect to the L2 inner product.
Numerical results confirming some features of the random wave model are given in [HR].
EγN cj = 0,
EγN cj ck =
1.15. Hyperbolic plane. It is useful to have one concrete example in mind, and we will use
the hyperbolic plane H or disc D as a running model. In the disc model D = {z ∈ C, |z| < 1},
4|dz|2
the hyperbolic metric has the form ds2 = (1−|z|
2 )2 . Its isometry group is G = P SU (1, 1); the
stabilizer of 0 is K ' SO(2) and thus D ' G/K. Without comment we will also identify D
with the upper half plane and G with P SL(2, R). In hyperbolic polar coordinates centered
at the origin 0, the (positive) Laplacian is the operator
∂2
∂
1
∂2
+
coth
r
+
.
∂r2
∂r sinh2 r ∂θ2
The distance on D induced by the Riemannian metric will be denoted dD . We denote the
volume form by d(z).
The unit tangent bundle SD of the hyperbolic disc D may be identifed with the unit
cosphere bundle S ∗ D by means of the metric. We further identify SD ≡ P SU (1, 1), since
P SU (1, 1) acts freely and transitively on SD. We identify a unit tangent vector (z, v) with
a group element g if g · (0, (1, 0)) = (z, v).
We denote by B = {z ∈ C, |z| = 1} the boundary at infinity of D. Then we can also
identify SD ≡ D × B. Here, we identify (z, b) ∈ D × B with the unit tangent vector (z, v),
where v ∈ Sz D is the vector tangent to the unique geodesic through z ending at b.
The geodesic flow g t on SD is defined by g t (z, v) = (γv (t), γv0 (t)) where γv (t) is the unit
speed geodesic with initial value (z, v). The space of geodesics is the quotient of SD by
the action of g t . Each geodesic has a forward endpoint b and a backward endpoint b0 in B,
hence the space of geodesics of D may be identified with B × B \ ∆, where ∆ denotes the
diagonal in B × B: To (b0 , b) ∈ B × B \ ∆ there corresponds a unique geodesic γb0 ,b whose
forward endpoint at infinity equals b and whose backward endpoint equals b0 . We then have
the identification
SD ≡ (B × B \ ∆) × R.
The choice of time parameter is defined – for instance – as follows: The point (b0 , b, 0) is by
definition the closest point to 0 on γb0 ,b and (b0 , b, t) denotes the point t units from (b, b0 , 0)
in signed distance towards b.
−4 =
12
STEVE ZELDITCH
1.15.1. Dynamics and group theory of G = SL(2, R). The generators of sl(2, R) are denoted
by






1 0
0 1
0 −1
, V = 
, W = 
.
H=
0 −1
1 0
1 0
We denote the associated one parameter subgroups by A, A− , K. We denote the raising/lowering operators for K-weights by
(23)
E + = H + iV,
E − = H − iV.
The Casimir operator is then given by 4 Ω = H 2 + V 2 − W 2 ; on K-invariant functions, the
Casimir operator acts as the laplacian 4. We also put




0 0
0 1
 , X− = 
,
X+ = 
1 0
0 0
and denote the associated subgroups by N, N− . In the identification SD ≡ P SL(2, R) the
geodesic flow is given by the right action g t (g) = gat of A, resp. the horocycle flow (hu )u∈R
is defined by hu (g) = gnu , where
 t/2



e
0
1 u
 , nu = 
.
at = 
−t/2
0 1
0
e
The closed orbits of the geodesic flow g t on Γ\G are denoted {γ} and are in one-to-one
correspondence with the conjugacy classes of hyperbolic elements of Γ. We denote by Gγ ,
respectively Γγ , the centralizer of γ in G, respectively Γ. The group Γγ is generated by an
element γ0 which is called a primitive hyperbolic geodesic. The length of γ is denoted Lγ > 0
and means that γ is conjugate, in G, to
 L /2

e γ
0
.
(24)
aγ = 
−Lγ /2
0
e
If γ = γ0k where γ0 is primitive, then we call Lγ0 the primitive length of the closed geodesic
γ.
1.15.2. Non-Euclidean Fourier analysis. Following [Hel], we denote by hz, bi the signed
distance to 0 of the horocycle through the points z ∈ D, b ∈ B. Equivalently,
ehz,bi =
1 − |z|2
= PD (z, b),
|z − b|2
where PD (z, b) is the Poisson kernel of the unit disc. We denote Lebesgue measure on B by
1
|db|. We then introduce the hyperbolic plane waves e( 2 +ir)hz,bi , the analogues of the Euclidean
plane waves eihx,ξi .
The non-Euclidean Fourier transform F : Cc D → C(B × R+ ) is defined by
Z
1
Fu(r, b) =
e( 2 −ir)hz,bi u(z)dV ol(z).
D
RECENT DEVELOPMENTS IN MATHEMATICAL QUANTUM CHAOS
13
The inverse Fourier transform is given by
Z Z
1
−1
F g(z) =
e( 2 +ir)hz,bi g(r, b)r tanh(2πr)dr|db|.
B
R
The integration measure is the Plancherel measure. F extends to an isometry F : L2 (D, dV ) →
L2 (B × R+ , r tanh(2πr)drdb).
√
1.15.3. Wave kernel on D and wave kernel on XΓ . The wave group U t = eit ∆ of the positive
Laplacian on D is thus given in terms of Fourier analysis by (distribution) integral
Z Z
1
1
1
2
t
U (z, w) =
ei( 4 +r )t e( 2 +ir)hz,bi e(− 2 +ir)hw,bi r tanh(2πr)dr|db|.
B
R
The integral over B may be eliminated by the stationary phase method to produce an integral
over R alone; it is the Hadamard parametrix (17) in this case.
Now let Γ ⊂ G be a discrete subgroup. In applications to quantum chaos we assume the
quotient XΓ is compact or of finite area.
The wave kernel on the quotient XΓ is obtained by automorphizing the wave kernel on D:
X
(25)
UΓt (z, w) =
UΓt (z, γw).
γ∈Γ
This representation is one of the inputs into the Selberg trace formula. The elements of Γ
are grouped into conjugacy classes [γ] ∈ [Γ]. The conjugacy classes correspond to closed
geodesics of XΓ , i.e. periodic orbits of g t on S ∗ XΓ .
Let
(26)
ΘΓ (T ) = #{[γ] : Lγ ≤ T }.
The prime geodesic theorem asserts that
ehtop T
,
T
where htop is the topological entropy of g t . In fact, htop = 1 for the hyperbolic case. The
exponential growth of the length spectrum reflects the exponential growth of the geodesic
flow.
The Ehrenfest time (18) is implicit in the exponential growth rate of ΘΓ (T ).
(27)
ΘΓ (T ) ∼
1.15.4. Representation theory of G and spectral theory of 4. Let Γ ⊂ G be a co-compact
discrete subgroup, and let us consider the automorphic eigenvalue problem on G/K:

 4ϕ = ( 14 + r2 )ϕ,
(28)

ϕ(γz) = ϕ(z) for all γ ∈ Γ and for all z.
The solutions are the eigenfunctions of the Laplacian on the compact surface XΓ = Γ\ G /K.
A standard notation for the eigenvalues is λ2 = s(1 − s) where s = 21 + ir.
Eigenfunctions of the Laplacian are closely connected with the representation theory of G
on G/Γ. We briefly describe the representation theory since the problems in quantum chaos
have analogues in the discrete series as well as the unitary prinicipal series.
14
STEVE ZELDITCH
In the compact case, we have the decomposition into irreducibles,
2
L (Γ\G) =
S
M
Cirj ⊕
j=1
∞
M
j=0
∞
M
Pirj ⊕
+
µΓ (m)Dm
m=2, m even
∞
M
⊕
−
µΓ (m)Dm
,
m=2,m even
where Cirj denotes the complementary series representation, respectively Pirj denotes the
unitary principal series representation, in which the Casimir operator −Ω = −(H 2 +V 2 −W 2 )
equals sj (1 − sj ) = 14 + rj2 . In the complementary series case, irj ∈ R while in the principal
series case irj ∈ iR+ . The irreducibles are indexed by their K-invariant vectors {ϕirj }, which
is assumed to be the given orthonormal basis of 4-eigenfunctions. Thus, the multiplicity of
Pirj is the same as the multiplicity of the corresponding eigenvalue of 4.
±
Further, Dm
denotes the holomorphic (respectively anti-holomorphic) discrete series representation with lowest (respectively highest) weight m, and µΓ (m) denotes its multiplicity;
it depends only on the genus of XΓ . We denote by ψm,j (j = 1, . . . , µΓ (m)) a choice of orµΓ (m) +
+
+
thonormal basis of the lowest weight vectors of µΓ (m)Dm
and write µΓ (m)Dm
= ⊕j=1
Dm,j
accordingly.
There is a a direct integral decomposition for co-finite subgroups such as Γ = P SL(2, Z)
or congruence subgroups. The non-compactness gives rise to a continuous spectral subspace
of Eisenstein series and a discrete spectral subspace of cuspidal eigenfunctions (which is only
known to be non-trivial in the case of arithmetic Γ).
1.15.5. Helgason Poisson formula. The Fourier transform of an L2 function on D is an L2
function on B ×R+ . There is an extension of the Fourier transform and the inversion formula
to tempered distributions. We only consider the case of Γ-automorphic eigenfunctions where
Γ is co-compact. One then has Fϕirj = dTirj (b) ⊗ δrj (r), where dTirj is a distribution on the
ideal boundary B. The inversion formula is Helgason’s Poisson formula,
Z
1
(29)
ϕir (z) =
e( 2 +ir)hz,bi Tir,ϕir (db),
B
1
( 1 +ir)
for all z ∈ D. The kernel e( 2 +ir)hz,bi = PD2
The distribution
(30)
Tir,ϕir (db) =
(z, b) is called the generalized Poisson kernel.
X
an (r)bn |db|.
n∈Z
is called the boundary value of ϕir and is obtained from the Fourier expansion
X
(31)
ϕir (z) =
an (r)Φr,n (z),
n∈Z
of ϕir in the disc model in terms of the generalized spherical functions,
X
1
Φr,n (z)bn , b ∈ B.
(32)
e( 2 +ir)hz,bi =
n∈Z
Equivalently, the Φr,n are the joint eigenfunctions of ∆ and of K.
RECENT DEVELOPMENTS IN MATHEMATICAL QUANTUM CHAOS
15
When ϕirj is a Γ-invariant eigenfunction, the boundary values Tirj (db) have the following
invariance property:
(33)
ϕirj (γz) = ϕirj (z)
1
1
=⇒ e( 2 +irj )hγz,γbi Tirj (dγb) = e( 2 +irj )hz,bi Tirj (db)
1
=⇒ Tirj (dγb) = e−( 2 +irj )hγ·0,γ·bi Tirj (db)
This follows by the identities
(34)
hg · z, g · bi = hz, bi + hg · 0, g · bi,
which implies
(35)
PD (gz, gb) |d(gb)| = PD (z, b) |db|.
An interesting heuristic related to Berry’s random wave hypothesis (see §1.14) is the
conjecture that the Fourier coefficients an (r) in (30) should behave like independent Gaussian
random variables of mean zero and variance one. This conjecture was stated explicitly and
tested numerically by Aurich-Steiner in [AS]. Related tests of the random wave model for
hyperbolic quotients are in Hejhal-Rackner [HR].
1
Otal [O] and Schmid [Sch] have shown that Tirj (db) is the derivative of a Hölder C 2
continuous function Firj on B. Since its zeroth Fourier coefficient is non-zero, Tirj (db) is not
literally the derivative of a periodic function, but it is the derivative of a function Firj on R
satisfying Firj (θ + 2π) = Firj (θ) + Cj for all θ ∈ R. Recall that for 0 ≤ δ ≤ 1 we say that
a 2π-periodic function F : R → C is δ-Hölder if |F (θ) − F (θ0 )| ≤ C|θ − θ0 |δ . The smallest
constant is denoted ||F ||δ and Λδ denotes the Banach space of δ-Hölder functions, up to
additive constants.
1.15.6. Boundary values and representation theory [Z3]. The distributions dTirj have a natural interpretation in representation theory. We define eirj ∈ D0 (Γ\P SL(2, R)) such that
(36)
1
e( 2 +irj )hz,bi Tirj (db) = eirj (z, b)P (z, b)db.
The distribution eirj is horocyclic-invariant and Γ-invariant. It may be expanded in a KFourier series,
X
eirj =
ϕirj ,n ,
n∈Z
and it is easily seen (cf. [Z2]) that ϕirj ,0 = ϕirj and that ϕirj ,n is obtained by applying the nth normalized raising or lowering operator (Maass operator) E ± = H ± iV to
ϕirj . More precisely, one applies (E ± )n and multiplies by the normalizing factor β2irj ,n =
1
.
(2irj +1±2n)···(2irj +1±2)
2. Weyl law and local Weyl law
We now return to the general case. A basic result in semi-classical asymptotics is Weyl’s
law on counting eigenvalues:
(37)
N (λ) = #{j : λj ≤ λ} =
|Bd |
V ol(M, g)λd + R(λ), where R(λ) = O(λd−1 ).
(2π)d
16
STEVE ZELDITCH
Here, |Bd | is the Euclidean volume of the unit ball and V ol(M, g) is the volume of M with
respect to the metric g. Weyl’s law says that
(38)
T rEλ ∼
V ol(|ξ|g ≤ λ)
,
(2π)d
P
where V ol is the symplectic volume measure relative to the natural symplectic form dj=1 dxj ∧
√
dξj on T ∗ M . Thus, the dimension of the space where H = ∆ is ≤ λ is asymptotically the
volume where its symbol |ξ|g ≤ λ.
The growth of the remainder term depends on the long time behavior of g t . It is sharp
on the standard sphere, where all geodesics are periodic. By a classical theorem due to
Duistermaat-Guillemin [D.G] (in the boundaryless case) and to Ivrii (in the boundary case),
R(λ) = o(λd−1 ), when the set of periodic geodesics has Liouville measure zero.
The remainder is then of small order than the derivative of the principal term, and one has
asymptotics in shorter intervals:
(39)
N ([λ, λ + 1]) = #{j : λj ∈ [λ, λ + 1]} = n
|Bd |
V ol(M, g)λd−1 + o(λd−1 ).
d
(2π)
Then the mean spacing between the eigenvalues in this interval is ∼ Cd V ol(M, g)−1 λ−(d−1) ,
where Cd is a constant depending on the dimension.
In the case of compact Riemannian manifolds of negative curvature, a sharper estimate of
the remainder is possible:
(40)
R(λ) = O(
λd−1
).
log λ
This remainder is proved using (25) and the exponential growth rate in (27). The estimate
(40) was proved by Selberg in the case of compact hyperbolic quotients and was generalized
to all compact Riemannian manifolds without conjugate points by Bérard [Be]. The logarithm in the remainder is a direct outcome of the fact that one can only use the geodesic
approximation to U t up to the Ehrenfest time TE (18).
This estimate has not been improved in fifty years, and there are no better results in
the constant curvature case than in the general negatively curved case. The remainder
estimate does not rule out the implausible-seeming scenario that in the interval [λ, λ + 1]
d−1
there are only log λ distinct eigenvalues with multiplicities λlog λ . In fact, such implausibly
large multiplicities do occur for a sparse set of ‘Planck constants’ in the analogous case of
quantizations of hyperbolic toral automorphisms (see [FNB]). The issue of possibly high
multiplicity or clustering of eigenvalues is important in Hassell’s scarring result as well as
the Anantharaman-Nonnenmacher results on entropies.
An important generalization is the local Weyl law concerning the traces T rAE(λ) where
A ∈ Ψm (M ). It asserts that
µZ
¶
X
1
(41)
hAϕj , ϕj i =
σA dxdξ λd + O(λd−1 ).
d
(2π)
B∗M
λ ≤λ
j
There is also a pointwise local Weyl law:
RECENT DEVELOPMENTS IN MATHEMATICAL QUANTUM CHAOS
(42)
X
λj ≤λ
|ϕj (x)|2 =
17
1
|B d |λd + R(λ, x),
d
(2π)
where R(λ, x) = O(λd−1 ) uniformly in x. When the periodic geodesics form a set of measure
zero in S ∗ M , the remainders are of order o(λd−1 ), and one could average over the shorter
interval [λ, λ + 1]. In the negatively curved case,
λd−1
).
log λ
Combining the Weyl and local Weyl law, we find the surface average of σA is a limit of traces:
Z
1
ω(A) :=
σA dµL
µL (S ∗ M ) S ∗ M
(43)
1 X
= lim
hAϕj , ϕj i
λ→∞ N (λ)
λ ≤λ
R(λ, x) = O(
j
Here, dµL is Liouville measure on S ∗ M (§1.1).
3. Invariant states defined by eigenfuntions and their quantum limits
When speaking of ‘states’ in quantum mechanics, one might refer to a normalized wave
function ψ or alternatively to the matrix elements ρψ (A) = hAψ, ψi of an observable in the
state. The latter use of ‘state’ is a standard notion in C* algebras, and is central in quantum
ergodicity and mixing. The states evidently have the properties: (i) ρψ (A∗ A) ≥ 0; (ii)ρψ (I) =
1; (iii)ρψ is continuous in the norm topology. The classical analogue is a probability measure,
viewed as a positive normalized linear functional on C(S ∗ M ). In particular, eigenfunctions
define states on Ψ0 (more correctly, its closure in the norm topology) as in (12), which we
repeat:
(44)
ρk (A) = hAϕk , ϕk i.
We could (and will) also consider ’transition amplitudes’ ρj,k (A) = hAϕj , ϕk i.
It is an immediate consequence of the fact that U t ϕj = eitλj ϕj that the diagonal states ρk
are invariant under the automorphism αt (14):
(45)
ρk (Ut AUt∗ ) = ρk (A).
In general, we denote by E the compact, convex set of states in the vector space of continuous
linear functionals on the closure of Ψ0 in its norm topology. We denote by
(46)
ER = {ρ ∈ E : ρ ◦ αt = ρ}
the compact convex set of invariant states. For simplicity of notation, we continue to denote
the closure by Ψ0 , and refer to [Z6] for a detailed exposition.
From a mathematical point of view, such the states ρk (A) are important because they
provide the simplest means of studying eigenfunctions: one tests eigenfunctions against
observables by studying the values ρk (A). Many standard inequalities in PDE (e.g. Carleman
estimates) have the form of testing eigenfunctions against well chosen observables. Of course,
the high eigenvalue asymptotics is not simple. One would like to know the behavior as
18
STEVE ZELDITCH
λj → ∞ (or ~ → 0 of the diagonal matrix elements hAϕj , ϕj i and the transition amplitudes
hAϕi , ϕj i between states for A ∈ Ψ0 (M ). One of the principal problems of quantum chaos
is the following:
Problem 3.1. Determine the set Q of ‘quantum limits’, i.e. weak* limit points of the
sequence of invariant eigenfunction stats {ρk } or equivalently of the Wigner distributions
{Wk }.
As will be illustrated below in simple examples, weak limits reflect the concentration and
oscillation properties of eigenfunctions.
Off-diagonal matrix elements
(47)
ρjk (A) = hAϕi , ϕj i
are also important as transition amplitudes between states. They no longer define states
since ρjk (I) = 0, are no longer positive, and are no longer invariant. Indeed, ρjk (Ut AUt∗ ) =
eit(λj −λk ) ρjk (A), so they are eigenvectors of the automorphism αt of (14). A sequence of such
matrix elements cannot have a weak limit unless the spectral gap λj − λk tends to a limit
τ ∈ R. Problem 3.1 has the following extension to off-diagonal elements:
Problem 3.2. Determine the set Qτ of ‘quantum limits’, i.e. weak* limit points of the
sequence {Wkj } of distributions on the classical phase space S ∗ M , defined by
Z
adWkj := hOp(a)ϕk , ϕj i
X
where λj − λk = τ + o(1) and where a ∈ C ∞ (S ∗ M ), or equivalently of the functionals ρjk .
3.1. Simplest properties of quantum limits. The first is that hKϕk , ϕk i → 0 for any
compact operator K. Indeed, any orthonormal basis such as {ϕk } tends to zero weakly in L2 .
Hence {Kϕk } tends to zero in L2 for any compact operator and in particular the diagonal
matrix elements tend to zero. It follows that for any A ∈ Ψ0 (M ), any limit of a sequence of
hAϕk , ϕk i is equally a limit of h(A + K)ϕk , ϕk i.
Any two choices of Op, i.e. of quantizations of homogeneous symbols (of order zero)
as pseudo-differential operators, are the same to leading order. Hence their difference is
compact. Since a negative order pseudo-differential operator is compact, Q is independent
of the definition of Op.
These properties do not use the fact that ϕj are eigenfunctions. The next property of the
ρk are consequences of the fact ρk ∈ ER .
Proposition 3.3. Q ⊂ MI , where MI is the convex set of invariant probability measures
for the geodesic flow. They are also time-reversal invariant.
To see this, we first observe that any weak * limit of of {ρk } is a positive linear functional
on symbols, which we identify with homogeneous functions of order zero on T ∗ M or with
smooth functions on S ∗ M . Indeed, any limit of hAϕk , ϕk i, is bounded by inf K ||A + K||
(the infimum taken over compact operators), and for any A ∈ Ψ0 , ||σA ||L∞ = inf K ||A + K||.
Hence any weak limit is bounded by a constant times ||σA ||L∞ and is therefore continuous
on C(S ∗ M ). It is a positive functional since each ρj is and hence any limit is a probability
measure. The invariance under g t follows from ρk ∈ ER by Egorov’s theorem: any limit of
RECENT DEVELOPMENTS IN MATHEMATICAL QUANTUM CHAOS
19
ρk (A) is a limit of ρk (Op(σA ◦ g t )) and hence the limit measure is g t invariant. Furthermore,
the limit measures are time-reversal invariant, i.e. invariant under (x, ξ) → (x, −ξ) since the
eigenfunctions are real-valued.
Problem 3.1 is thus to identify which invariant measures in MI show up as weak limits of
the functionals ρk or equivalently of the Wigner distributions dWk . The problem is that MI
can be a very large set. Examples of invariant probability measures for the geodesic flow
include:
(1) Normalized Liouville measure dµL . In fact, the functional ω of (43) is also a state
on Ψ0 for the reason explained above. A subsequence {ϕjk } of eigenfunctions is
considered diffuse if ρjk → ω.
R
(2) A periodic orbit measure µγ defined by µγ (A) = L1γ γ σA ds where Lγ is the length
of γ. A sequence of eigenfunctions for which ρkj → µγ obviously concentrates (or
strongly ‘scars’) on the closed geodesic.
P
(3) A finite convex combination M
j=1 cj dµγj of periodic orbit measures.
P
(4) A mixed measure such as 21 dµL + 12 M
j=1 cj dµγj
(5) A delta-function along an invariant Lagrangian manifold Λ ⊂ S ∗ M . The associated
eigenfunctions are viewed as localizing along Λ.
(6) There are many additional kinds of singular measures in the Anosov case.
Thus, the constraint in the Proposition is far from enough to determine Q.
To pin down Q, it is necessary to find more constraints on the states ρk and their weak*
limits. We must use the quantum mechanics of ρk to pin down the possible classical limits.
What possible additional constraints are there on the ρk ? To date, only two are known.
• Further symmetries of the ϕj . In general there are none. But in special cases they
exist. The most significant case is that of Hecke eigenfunctions, which carry an
infinite number of further symmetries. See §9 and [LIND, Sound1].
• Entropies of limit measures. Lower bounds were obtained by N. Anantharaman and
S. Nonnemacher [A, AN] in the case of (M, g) with Anosov geodesic flow (see also
[ANK, Riv].)
These constraints are of a very different nature. The Hecke symmetries are special to
arithmetic hyperbolic quotients, where one supplement the geodesic flow–wave group connection with the Hecke symmetries and connections to number theory. The entropy bounds
are very general and only use the geodesic flow–wave group connection.
3.2. Ergodic sequences of eigenfunctions. A subsequence {ϕjk } of eigenfunctions is
called ergodic if the only weak * limit of the sequence of ρjk is dµL or equivalently the
Liouville state ω. If dWkj → ω then in particular, we have
Z
1
V ol(E)
|ϕkj (x)|2 dV ol →
V ol(M ) E
V ol(M )
for any measurable set E whose boundary has measure zero. In the interpretation of
|ϕkj (x)|2 dV ol as the probability density of finding a particle of energy λ2k at x, this says
that the sequence of probabilities tends to uniform measure.
20
STEVE ZELDITCH
However, Wkj → ω is much stronger since it says that the eigenfunctions become diffuse
on the energy surface S ∗ M and not just on the configuration space M . One can quantize
characteristic functions 1E of open sets in S ∗ M whose boundaries have measure zero. Then
hOp(1E )ϕj , ϕj i = the amplitude that the particle in energy state λ2j lies in E.
For an ergodic sequence of eigenfunctions,
hOp(1E )ϕjk , ϕjk i →
µL (E)
,
µL (S ∗ M )
so that the particle becomes diffuse, i.e. uniformly distributed on S ∗ M . This is the quantum
analogue of the property of uniform distribution of typical geodesics of ergodic geodesic flows
(Birkhoff’s ergodic theorem).
3.3. QUE. The Laplacian ∆ or (M, g) is said to be QUE (quantum uniquely ergodic) if
Q = {µL }, i.e. the only quantum limit measure for any orthonormal basis of eigenfunctions
is Liouville measure.
Conjecture 3.4. (Rudnick-Sarnak, [RS]) Let (M, g) be a negatively curved manifold. Then
∆ is QUE.
In [RS], the case of arithmetic manifolds is investigated and the role of the Hecke operators
is clarified and exploited. In particular the arithmetic QUE Conjecture refers to the limits of
Hecke eigen-states, i.e. eigenfunctions of the arithmetic symmetries called Hecke operators.
As reviewed in §9, in this case the conjecture in all its forms is more or less completely
solved. E. Lindenstrauss [LIND] (together with the recent final step in the noncompact
case by Soundararajan [Sound1]) proves this arithmetic QUE for arithmetic surfaces. If
the multiplicity of the eigenvalues are uniformly bounded then one can deduce full QUE
from arithmetic QUE, i.e. QUE for any orthonormal basis of eigenfunctions (see §9). Such
uniform bounds on multiplicities are however far out of the range of current technology .
The analogue of arithmetic QUE for Hecek holomorphic forms on noncompact arithmetic
surfaces has recently been settled by Holowinsky and Soundararajan [Hol, HS]. In this case,
the Hecke condition cannot be dropped due to the high multiplicity of such forms. Their
methods are entirely arithmetical and we wont discuss them further here.
Although we are not discussing quantum cat maps in detail, it should be emphasized that
quantizations of hyperbolic (Anosov) symplectic maps of the torus) are not QUE. For a
sparse sequence of Planck constants ~k , there exist eigenfunctions of the quantum cat map
which partly scar on a hyperbolic fixed point (see Faure-Nonnenmacher-de Bièvre [FNB]).
The multiplicities of the corresponding eigenvalues are of order ~−1
k /| log ~|. It is unknown if
anything analogous can occur in the Riemannian setting, but as yet there is nothing to rule
it out.
As pointed out in [Z10], QUE would follow if one could prove that quantum limits were
invariant under a uniquely ergodic flow such as the horocycle flow of a compact hyperbolic
quotient. A problem in trying to use this approach is that the horocycle flow is not Hamiltonian with respect to the standard symplectic form on T ∗ XΓ , i.e. it cannot be quantized.
It is however Hamiltonian with respect to a modified symplectic structure. The modified
symplectic structure corresponds to letting the weight of automorphic forms vary with the
∆ eigenvalue. As a result, one does get large sequences of automorphic forms which are
RECENT DEVELOPMENTS IN MATHEMATICAL QUANTUM CHAOS
21
Liouville distributed. But QUE in the standard sense refers to sequences with fixed weight.
It turns out that the weight has to grow so quickly for the QUE sequence that one cannot
seem to relate the QUE result to quantum limits of forms with fixed weight [Z10].
3.4. Simplest example: S 1 . The only computable example of ergodic eigenfunctions is
the sequence of normalized eigenfunctions 2 sin kπx, 2 cos kψx on S 1 . Note that they are real
valued; the exponentials eikx are not quantum ergodic.
The energy surface of T ∗ S 1 = S 1 × Rξ is the pair of circles |ξ| = ±1. The geodesic flow has
two invariant sets of positive measure (the two components), but the flow is time-reversal
invariant under (x, ξ) → (x − ξ) and the quotient flow is ergodic. The real eigenfunctions are
invariant under complex conjugation and therefore the quotient quantum system is ergodic.
Quantum ergodicity in this case amounts to
Z
Z 2π
1 2π
1
2
V (x)(sin kx) dx →
V dx.
π 0
2π 0
This is obvious by writing sin kx in terms of exponentials and using the Riemann-Lebesgue
Lemma.
This simple example illustrates an important aspect of quantum limits: They are weak limits which owe to the fast oscillation of the eigenfunction squares. In the limit the oscillating
functions tend to their mean values in the weak sense.
In particular, it illustrates why we only consider squares and not other powers of eigenfunctions: weak limits are not preserved under non-linear functionals such as powers. The
study of Lp norms of chaotic eigenfunctions is very difficult (there are Iwaniec-Sarnak for
Hecke eigenfunctions).
3.5. Example of quantum limits: the flat torus. The only examples where one can
compute quantum limits directly are the completely integrable ones such as the standard
sphere, torus or symmetric spaces. These examples of course lie at the opposite extreme from
chaotic or ergodic dynamics. We use the simplest one, the flat torus Rn /Zn , to illustrate the
definition of weak* limits.
An orthonormal basis of eigenfunctions is furnished by the standard exponentials e2πihk,xi
with k ∈ Zn . Obviously, |e2πihk,xi |2 = 1, so the eigenfunctions are diffuse in configuration
space. But they are far from diffuse in phase space. For any pseudodifferential operator,
Ae2πihk,xi = a(x, k)e2πihk,xi where a(x, k) is the complete symbol. Thus,
Z
Z
k
2πihk,xi 2πihk,xi
hAe
,e
i=
a(x, k)dx ∼
σA (x, )dx.
|k|
Rn /Zn
Rn /Zn
Thus, the Wigner distribution is δξ−k . A subsequence e2πihkj ,xi of eigenfunctions has a weak
k
limit if and only if |kjj | tends to a limit vector ξ0 in the unit sphere in Rn . In this case, the
R
associated weak* limit is Rn /Zn σA (x, ξ0 )dx, i.e. the delta-function on the invariant torus
Tξ0 ⊂ S ∗ M defined by the constant momentum condition ξ = ξ0 . The eigenfunctions are
said to localize under this invariant torus for g t .
The invariant torus is a Lagrangian submanifold of T ∗ Rn /Zn , i.e. a submanifold of dimension n on which the standard symplectic form dx ∧ dξ restricts to zero. The exponentials
i
are special cases of WKB or Lagrangian states a~ e ~ S , where a~ is a semi-classical symbol
22
STEVE ZELDITCH
a~ ∼ aj ~j . The associated Lagrangian submanifold is the graph (x, dS(x)) of S. Thus, hk, xi
k
generates the Lagrangian submanifold ξ = |k|
.
In general, one says that a sequence {ϕjk } of eigenfunctions concentrates microlocally on
a Lagrangian submanifold Λ ⊂ S ∗ M if
Z
(48)
hAϕjk , ϕjk i →
σA dν,
Λ
for some probability measure ν on Λ. Necessarily, Λ is invariant under g t .
3.6. Hyperbolic case. Using the Helgason Poisson integral formula, the Wigner distributions can be expressed in terms of the dTirh and eirj as follows. As in [Z3], we define the
hyperbolic calculus of pseudo-differential operators Op(a) on D by
1
1
Op(a)e( 2 +ir)hz,bi = a(z, b, r)e( 2 +ir)hz,bi .
We assume that the complete symbol a is a polyhomogeneous function of r in the classical
sense that
∞
X
a(z, b, r) ∼
aj (z, b)r−j+m
j=0
for some m (called its order). By asymptotics is meant that
a(z, b, r) −
R
X
aj (z, b)r−j+m ∈ S m−R−1
j=0
where σ ∈ S k if sup(1 + r)j−k |Dzα Dbβ Drj σ(z, b, r)| < +∞ for all compact set and for all α, β, j.
The non-Euclidean Fourier inversion formula then extends the definition of Op(a) to
∞
Cc (D):
Z Z
1
Op(a)u(z) =
a(z, b, r)e( 2 +ir)hz,bi Fu(r, b)r tanh(2πr)dr|db|.
B
R
A key property of Op is that Op(a) commutes with the action of an element γ ∈ G
(Tγ u(z) = u(γz)) if and only if a(γz, γb, r) = a(z, b, r). Γ-equivariant pseudodifferential
operators then define operators on the quotient XΓ .
By the Helgason-Poisson formula one has a relative explicit formula for the Wigner distributions Wirj ∈ D0 (S ∗ XΓ )) defined by
Z
(49)
ha, Wirj i =
a(g)Wirj (dg) := hOp(a)ϕirj , ϕirj iL2 (XΓ ) , a ∈ C ∞ (S ∗ XΓ ).
S ∗ XΓ
Equivalently we have
(50)
Wirj = ϕirj eirj .
Thus, the Wigner distributions are far more diffuse in phase space than in the case of the flat
torus. But this does not rule out that their weak* limits could have a singular concentration.
RECENT DEVELOPMENTS IN MATHEMATICAL QUANTUM CHAOS
23
3.6.1. Patterson-Sullivan distributions. Egorov’s theorem implies that Wigner distributions
tend to invariant measures for the geodesic flow. The question arises whether there exist g t invariant distributions constructed from eigenfunctions which are asymptotic to the Wigner
distributions. In [AZ] such distributions were constructed on hyperbolic surfaces and termed
“Patterson-Sullivan distributions” by analogy with their construction of boundary measures
associated to ground states on infinite volume hyperbolic manifolds.
Definition: The Patterson-Sullivan distribution associated to a real eigenfunction ϕirj is
the distribution on B × B \ ∆ defined by
psirj (db0 , db) :=
Tirj (db)Tirj (db0 )
|b − b0 |1+2irj
If ϕirj is Γ-automorphic, then psirj (db0 , db) is Γ-invariant and time reversal invariant.
To obtain a g t -invariant distribution on S ∗ XΓ , we tensor psirj with dt. We then normalize
by dividing by the integral against 1. The result is an invariant distribution PˆS irj for g t
constructed as a quadratic expression in the eigenfunctions. In [AZ] Theorem 1.1, it is proved
(theorem that
Z
Z
S ∗ XΓ
aPˆS irj =
S ∗ XΓ
aWirj + O(rj−1 ).
Hence the quantum limits problem is equally one of determining the weak* limits of the
Patterson-Sullivan distributions. It is shown in [AZ] that they are residues of dynamical
L-functions and hence have a purely classical definition.
In fact, there is an explicit intertwining operator Lrj mapping P Sirj → Wirj and we have
(51)
ha, Wirj i = ha, PˆS irj i + rj−1 hL2 (a), PˆS irj i + O(rj−2 ).
4. Quantum ergodicity and mixing of eigenfunctions
In this section, we review a basic result on quantum ergodicity. We assume that the
geodesic flow of (M, g) is ergodic. Ergodicity of g t means that Liouville measure dµL is an
ergodic measure for g t on S ∗ M , i.e. an extreme point of MI . That is, there any g t -invariant
set has Liouville measure zero or one. Ergodicity is a spectral property of the operator V t of
(15) on L2 (S ∗ M, dµL ), namely that V t has 1 as an eigenvalue of multiplicity one. That is, the
only invariant L2 functions (with respect to Liouville measure) are the constant functions.
In this case, there is a general result which originated in the work of A. I. Schnirelman and
was developed into the following theorem by S. Zelditch, Y. Colin de Verdière on manifolds
without boundary and by P. Gérard-E. Leichtnam and S. Zelditch-M. Zworski on manifolds
with boundary.
Theorem 4.1. Let (M, g) be a compact Riemannian manifold (possibly with boundary), and
let {λj , ϕj } be the spectral data of its Laplacian ∆. Then the geodesic flow Gt is ergodic on
(S ∗ M, dµL ) if and only if, for every A ∈ Ψo (M ), we have:
P
• (i) limλ→∞ N 1(λ) λj ≤λ |(Aϕj , ϕj ) − ω(A)|2 = 0.
P
• (ii) (∀²)(∃δ) lim supλ→∞ N 1(λ) j6=k:λj ,λk ≤λ |(Aϕj , ϕk )|2 < ²
|λj −λk |<δ
24
STEVE ZELDITCH
The diagonal result may be interpreted as a variance result for the local Weyl law. Since all
the terms are positive, the asymptotic is equivalent to the existence of a s subsequence {ϕjk }
of eigenfunctions whose indices jk have counting density one for which hAϕjk , ϕjk i → ω(A)
for any A ∈ Ψ0 (M ). As above, such a sequence of eigenfunctions is called ergodic. One can
sharpen the results by averaging over eigenvalues in the shorter interval [λ, λ + 1] rather than
in [0, λ].
The off-diagonal statement was proved in [Z9] and the fact that its proof can be reversed
to prove the converse direction was observed by Sunada in [Su]. A generalization to finite
area hyperbolic surfaces is in [Z8].
The first statement (i) is essentially a convexity result. It remains true if one replaces the
square by any convex function F on the spectrum of A,
1 X
F (hAϕk , ϕk i − ω(A)) → 0.
(52)
N (E) λ ≤E
j
4.1. Quantum ergodicity in terms of operator time and space averages. The diagonal variance asymptotics may be interpreted as a relation between operator time and space
averages.
Definition Let A ∈ Ψ0 be an observable and define its time average to be:
hAi := lim hAiT ,
T →∞
where
1
hAiT :=
2T
Z
T
U t AU −t dt
−T
and its space average to be scalar operator
ω(A) · I
Then Theorem 4.1 (1) is (almost) equivalent to,
(53)
hAi = ω(A)I + K,
where
lim ωλ (K ∗ K) → 0,
λ→∞
where ωλ (A) = N 1(λ) T rE(λ)A. Thus, the time average equals the space average plus a term K
which is semi-classically small in the sense that its Hilbert-Schmidt norm square ||Eλ K||2HS
in the span of the eigenfunctions of eigenvalue ≤ λ is o(N (λ)).
This is not exactly equivalent to Theorem 4.1 (1) since it is independent of the choice of
orthonormal basis, while the previous result depends on the choice of basis. However, when
all eigenvalues have multiplicity
one, then the two are equivalent. To see the equivalence, note
√
that hAi commutes with ∆ and hence is diagonal in the basis {ϕj } of joint eigenfunctions
of hAi and of Ut . Hence K is the diagonal matrix with entries hAϕk , ϕk i − ω(A). The
condition is therefore equivalent to
1 X
lim
|hAϕk , ϕk i − ω(A)|2 = 0.
E→∞ N (E)
λ ≤E
j
RECENT DEVELOPMENTS IN MATHEMATICAL QUANTUM CHAOS
25
4.2. Heuristic proof of Theorem 4.1 (i). There is a simple picture of eigenfunction states
which makes Theorem 4.1 seem obvious. Justifying the picture is more difficult than the
formal proof below but the reader may find it illuminating and convincing.
First, one should re-formulate the ergodicity of g t as a property of the Liouville measure
dµL : ergodicity is equivalent to the statement dµL is an extreme point of the compact convex
set MI . Moreover, it implies that the Liouville state ω on Ψ0 (M ) is an extreme point of the
compact convex set ER of invariant states for αt of (14); see [Ru] for background.
But the
P
1
local Weyl law says that ω is also the limit of the convex combination N (E)
λj ≤E ρj . An
extreme point cannot be written as a convex combination of other states unless all the states
in the combination are equal to it. In our case, ω is only a limit of convex combinations so
it need not (and does not) equal each term. However, almost all terms in the sequence must
tend to ω, and that is equivalent to (1).
One could make this argument rigorous by considering whether Liouville measure is an
exposed point of EI and MI . Namely, is there a linear functionalP
Λ which is equal to zero
1
at ω and is < 0 everywhere else on EI ? If so, the fact that N (E)
λj ≤E Λ(ρj ) → 0 implies
that Λ(ρj ) → 0 for a subsequence of density one. For one gets an obvious contradiction if
Λ(ρjk ) ≤ −² < 0 for some ² > 0 and a subsequence of positive density. But then ρjk → ω
since ω is the unique state with Λ(ρ) = 0.
In [J] it is proved that Liouville measure (or any ergodic measure) is exposed in MI . It is
stated in the following form: For any ergodic invariant probability measure µ, there exists
a continuous function f on S ∗ M so that µ is the unique f -maximizing measure in the sense
that
½Z
¾
Z
f dm : m ∈ MI .
f dµ = sup
To complete the proof, one would need to show that the extreme point ω is exposed in EI
for the C* algebra defined by the norm-closure of Ψ0 (M ).
4.3. Sketch of Proof of Theorem 4.1 (i). We now sketch the proof of (52). By time
averaging, we have
X
X
F (hhAiT − ω(A)ϕk , ϕk i).
F (hAϕk , ϕk i − ω(A)) =
(54)
λj ≤E
λj ≤E
We then apply the Peierls–Bogoliubov inequality
n
X
F ((Bϕj , ϕj )) ≤ Tr F (B)
j=1
with B = ΠE [hAiT − ω(A)]ΠE to get:
X
(55)
F (hhAiT − ω(A)ϕk , ϕk i) ≤ Tr F (ΠE [hAiT − ω(A)]ΠE ).
λj ≤E
Here, ΠE is the spectral projection for Ĥ corresponding to the interval [0, E]. By the
Berezin inequality (if F (0) = 0):
1
Tr F (ΠE [hAiT
N (E)
− ω(A)]ΠE ) ≤
1
Tr ΠE F ([hAiT
N (E)
− ω(A)])ΠE
= ωE (ϕ(hAiT − ω(A))).
26
STEVE ZELDITCH
As long as F is smooth, F (hAiT − ω(A)) is a pseudodifferential operator of order zero with
principal symbol F (hσA iT − ω(A)). By the assumption that ωE → ω we get
Z
1 X
lim
F (hAϕk , ϕk i − ω(A)) ≤
F (hσA iT − ω(A))dµL .
E→∞ N (E)
{H=1}
λ ≤E
j
As T → ∞ the right side approaches ϕ(0) by the dominated convergence theorem and by
Birkhoff’s ergodic theorem. Since the left hand side is independent of T , this implies that
1 X
lim
F (hAϕk , ϕk i − ω(A)) = 0
E→∞ N (E)
λ ≤E
j
for any smooth convex F on Spec(A) with F (0) = 0.
¤
This proof can only be used directly for scalar Laplacians on manifolds without boundary,
but it still works as a template in more involved situations. For instance, on manifolds
with boundary, conjugation by the wave group is not a true automorphism of the observable
algebra. In quantum ergodic restriction theorems (see §6), the appropriate conjugation is an
endomorphism but not an automorphism. Or when ∆ has continuous spectrum (as in finite
area hyperbolic surfaces), one must adapt the proof to states which are not L2 -normalized
[Z8].
4.4. QUE in terms of time and space averages. The quantum unique ergodicity problem (the term is due to Rudnick-Sarnak [RS]) is the following:
Problem 4.2. Suppose the geodesic flow g t of (M, g) is ergodic on S ∗ M . Is the operator K
in
hAi = ω(A) + K
√
a compact operator? Equivalently is Q = {dµL }? In this case, ∆ is said to be QUE
(quantum uniquely ergodic)
Compactness of K implies that hKϕk , ϕk i → 0, hence hAϕk , ϕk i → ω(A) along the entire
sequence.
Rudnick-Sarnak conjectured that ∆ of negatively curved manifolds are QUE, i.e. that for
any orthonormal basis of eigenfunctions, the Liouville measure is the only quantum limit
[RS].
4.5. Converse QE. So far we have not mentioned Theorem 4.1 (2). An interesting open
problem is the extent to which (2) is actually necessary for the equivalence to classical
ergodicity.
√
Problem 4.3. Suppose that ∆ is quantum ergodic in the sense that (1) holds in Theorem
4.1. What are the properties of the geodesic flow g t . Is it ergodic (in the generic case)?
In the larger class of Schrödinger operators, there is a simple example of a Hamiltonian
system which is quantum ergodic but not classically ergodic: namely, a Schrödinger operator
with a symmetric double well potential W . That is, W is a W shaped potential with two
wells and a Z2 symmetry exchanging the wells. The low energy levels consist of two connected
components interchanged by the symmetry, and hence the classical Hamiltonian flow is not
d2
ergodic. However, all eigenfunctions of the Schrödinger operator − dx
2 + W are either even
RECENT DEVELOPMENTS IN MATHEMATICAL QUANTUM CHAOS
27
or odd and thus have the same mass in both wells. It is easy to see that the quantum
Hamiltonian is quantum ergodic.
Recently, B. Gutkin [Gut] has given a two dimensional example of a domain with boundary
which is quantum ergodic but not classically ergodic and which is a two dimensional analogue
of a double well potential. The domain is a so-called hippodrome (race-track) stadium.
Similarly to the double well potential, there are two invariant sets interchanged by a Z2
symmetry. They correspond to the two orientations with which the race could occur. Hence
the classical billiard flow on the domain is not ergodic. After dividing by the Z2 symmetry
the hippodrome has ergodic billiards, hence by Theorem 4.1, the quotient domain is quantum
ergodic. But the The eigenfunctions are again either even or odd. Hence the hippodrome is
quantum ergodic but not classically ergodic.
Little is known about converse quantum ergodicity in the abscence of symmetry. It is
known that if there exists an open set in S ∗ M filled by periodic orbits, then the Laplacian
cannot be quantum ergodic (see [MOZ] for recent results and references). But it is not even
known at this time whether KAM systems, which have Cantor-like invariant sets of positive
measure, are not quantum ergodic. It is known that there exist a positive proportion of
approximate eigenfunctions (quasi-modes) which localize on the invariant tori, but it has
not been proved that a positive proportion of actual eigenfunctions have this localization
property.
4.6. Quantum weak mixing. There are parallel results on quantizations of weak-mixing
geodesic flows. We recall that the geodesic flow of (M, g) is weak mixing if the operator V t
has purely continuous spectrum on the orthogonal complement of the constant functions in
L2 (S ∗ M, dµL ).
Theorem 4.4. [Z8] The geodesic flow g t of (M, g) is weak mixing if and only if the conditions
(1)-(2) of Theorem 4.1 hold and additionally, for any A ∈ Ψo (M ),
X
1
(∀²)(∃δ) lim sup
|(Aϕj , ϕk )|2 < ²
(∀τ ∈ R)
N
(λ)
λ→∞
j6=k:λ ,λ ≤λ
j k
|λj −λk −τ |<δ
The restriction j 6= k is of course redundant unless τ = 0, in which case the statement
coincides with quantum ergodicity. This result follows from the general asymptotic formula,
valid for any compact Riemannian manifold (M, g), that
¯
¯2
P
1
2 ¯ sin T (λi −λj −τ ) ¯
|hAϕ
,
ϕ
i|
¯ T (λi −λj −τ ) ¯
i
j
i6=j,λi ,λj ≤λ
N (λ)
(56)
R T itτ
1
∼ || 2T
e Vt (σA )||22 − | sinT Tτ τ |2 ω(A)2 .
−T
In the case of weak-mixing geodesic flows, the right hand side → 0 as T → ∞. As with
diagonal sums, the sharper result is true where one averages over the short intervals [λ, λ+1].
Theorem 4.4 is based on expressing the spectral measures of the geodesic flow in terms of
matrix elements. The main limit formula is:
Z
τ +ε
(57)
τ −ε
1
λ→∞ N (λ)
dµσA := lim
X
i,j: λj ≤λ, |λi −λj −τ |<ε
|hAϕi , ϕj i|2 ,
28
STEVE ZELDITCH
where dµσA is the spectral measure for the geodesic flow corresponding to the principal
symbol of A, σA ∈ C ∞ (S ∗ M, dµL ). Recall that the spectral measure of V t corresponding to
f ∈ L2 is the measure dµf defined by
Z
t
hV f, f iL2 (S ∗ M ) =
eit τ dµf (τ ) .
R
4.7. Evolution of Lagrangian states. In this section, we briefly review results on evolution of Lagrangian states and coherent states. We follow in particular the article of R.
Schubert [Schu3].
i
A simple Lagrangian orP
WKB state has the form ψ~ (x) = a(~, x)e ~ S(x) where a(~, x) is a
j
semi-classical symbol a ∼ ∞
j=0 ~ aj (x). The phase S generates the Lagrangian submanifold
(x, dS(x)) ⊂ T ∗ M .
It is proved in Theorem 1 of [Schu3] that if g t is Anosov and if Λ is transversal to the
stable foliation W s (except on a set of codimension one), then there exists C, τ > 0 so that
for every smooth density on Λ and every smooth function a ∈ C ∞ (S ∗ M ), the Lagrangian
state ψ with symbol σψ satisfies,
¯
¯
Z
Z
¯ t
¯
t
2¯
Γ|t|
−tτ
¯
(58)
hU
ψ,
AU
ψi
−
σ
dµ
|σ
|
A
L
ψ
¯
¯ ≤ Che + ce .
S∗M
Λ
In order that the right side tends to zero as ~ → 0, t → ∞ it is necessary and sufficient that
t≤
1−²
| log ~|.
Γ
5. Concentration of eigenfunctions around hyperbolic closed geodesics
As mentioned above, the quantum ergodicity Theorem 4.1 leaves open the possible existence of a sparse (zero density) subsequence of eigenfunctions which ‘weakly scar’ on a
hyperbolic orbit γ in the sense that its quantum limit ν0 contains a non-trivial multiple
of the periodic orbit measure cµγ as a non-zero ergodic component. The Anantharaman
entropy bound shows that when (M, g) is Anosov, there cannot exist such a sequence of
eigenfuntions (or even quasi-modes) which tend to µγ itself, but a quantum limit could have
the form c1 µγ + c2 µL for certain c1 , c2 satisfying c1 + c2 = 1. The question we address in this
section is the possible mass profile of such a scarring eigenfunction in a neighborhood of γ.
More precisely, how much mass does ϕλ have in a shrinking ~1/2−² tube around a hyperbolic
closed geodesic? This question will surface again in §11.21.
We first note that the existence of a quantum limit of the form c1 µγ + c2 µL for (M, g)
with Anosov geodesi flow is not so implausible. Analogous eigenfunction sequences do exist
for the so-called quantum cat map [FNB, FN]. And exceptional sequences ‘scarring’ on a
certain 1-parameter family of periodic orbits exists for the Bunimovich stadium [Has]. At
this time, there are no known examples of sequences of eigenfunctions or quasi-modes for
(M, g) with ergodic geodesic flow that ‘weakly scar’ along a hyperbolic closed orbit γ and
no results prohibiting them.
For simplicity, assume that (M, g) is a Riemannian manifold of dimension 2 with a closed
geodesic γ of length L. We assume γ is an embedded (non self-intersecting) curve. If ϕj is
RECENT DEVELOPMENTS IN MATHEMATICAL QUANTUM CHAOS
29
a Laplace eigenfunction, we define the mass profile of ϕj near γ to be the function
Z
M (ϕj )(r) =
|ϕj |2 dS,
d(x,γ)=r
where dS ∧ dr = dV .
Before considering possible mass profiles of eigenfunctions scarring on hyperbolic closed
geodesics, let us recall the opposite and much better known case of scarring of Gaussian
beams along elliptic closed geodesics on surfaces [Ra, Ra2, B.B]. When γ is an elliptic
closed geodesic, then there always exists a sequence of quasi-modes (Gaussian beams) which
concentrates on γ. As the
suggests, Gaussian beams ψλ oscillate like eiλs along γ
√ name
and resemble Gaussians λe−λhA(s)y,yi (for some positive symmetric matrix A(s)) in the
transverse direction with height λ1/4 and concentrated in a √1λ tube around γ. Thus, the
mass profile is a Gaussian probability measure with mean zero and variance λ−1/2 . The local
model for such quasi-modes is that of a harmonic oscillator in the fibers of Nγ . One can
construct the quasi-mode so that it is of infinite order. Of course, stable elliptic orbits do
not exist when (M, g) has ergodic geodesic flow (by the KAM theorem).
Now consider hyperbolic closed orbits. It was pointed out by Duistermaat [Dui] (section 1.5) that one cannot construct analogous quasi-modes associated to hyperbolic closed
geodesics as Lagrangian states. The stable/unstable manifolds Λ± of γ, containing the
geodesics which spiral in towards γ, are invariant Lagrangian submanifolds, but the only
invariant half-density on the Lagrangians is the ‘delta’-density on the closed geodesic.
Further, there are apriori limitations on the degree to which eigenfunctions sequences can
concentrate around hyperbolic closed geodesic on any (M, g).
Theorem 5.1. [BZ, Chr, CVP] Let (M, g) be a compact Riemannian manifold, and let γ
be a hyperbolic closed geodesic. Let U be any tubular neighborhood of γ in M . Then for any
eigenfunction ϕλ , there exists a constant C depending only on U such that
Z
C
|ϕλ |2 dVg ≥
||ϕλ ||2L2 .
log
λ
M \U
More generally, let A ∈ Ψ0 (M ) be a pseudo-differential orbit whose symbol equals one in
a neighborhood of γ in Sg∗ M and equals zero outside another neighborhood. Then for any
C
eigenfunction ϕλ ||(I − A)ϕλ ||L2 ≥ √log
||ϕλ ||L2 .
λ
This allows scarring sequences along a hyperbolic orbit to occur, it just limits the rate
at which the mass concentrates near γ. It implies that the mass profile of a sequence of
eigenfunctions concentrating on a hyperbolic close geodesic has “long tails”, i.e. there is a
fairly large amount of mass far away from the geodesic, although sequences with the quantum
limit µγ must tend to zero outside of any tube around the closed geodesic. Note that this
result makes no dynamical hypotheses. It applies equally to (M, g) with integrable geodesic
flow and to those with Anosov geodesic flow. To the author’s knowledge, there do not exist
more precise results in the Anosov case.
An obvious question at this point is whether there are any examples of (M, g), with any
type of geodesic flow, possessing a sequence of eigenfunctions scarring on a closed hyperbolic
orbit. The answer to this question is ‘yes’. It is simple to see that such eigenfunctions exist
in the opposite extreme of completely integrable systems, for instances surfaces of revolution
30
STEVE ZELDITCH
like peanuts with hyperbolic waists. Examples include joint eigenfunctions of the square root
of the Laplacian and rotation on surfaces of revolution with a hyperbolic waist. A truncated
hyperbolic cylinder is another example studied in [CVP]. In this case, the joint spectrum
∂
fills out the image of the moment map (pθ , |ξ|) : T ∗ M → R2 , where pθ (x, ξ) = hξ, ∂θ
i is the
angular momentum. At critical distances to the axis of rotation, the lattitude circles are
closed geodesics and the level set of the moment map becomes singular. If the surface is
shaped like a peanut, the waist is a hyperbolic closed geodesic. Joint eigenfunctions whose
joint eigenvalues are asymptotic to the singular levels of the moment always exist. The
modes concentrate on the level sets of the moment map, and in fact they concentrate on the
hyperbolic closed geodesic.
5.1. Mass concentration of special eigenfunctions on hyperbolic orbits in the
quantum integrable case. The mass profile of scarring eigenfunctions near a hyperbolic
in the completely integrable case is studied in [CVP] on tubes of fixed radius and in [NV, TZ2]
on tubes of shrinking radius. Let γ ⊂ S ∗ M be a closed hyperbolic geodesic of an (M, g) with
completely integrable geodesic flow and for which ∆g is quantum integrable (i.e. commutes
with a maximal set of pseudo-differential operators; see [TZ2] for background). We then
consider joint eigenfunctions ∆g and of these operators. It is known (see e.g. [TZ2], Lemma
6) that there exists a special sequence of eigenfunctions concentrating on the momentum
level set of γ. We will call them (in these notes) the γ-sequence.
Assume for simplicity that the moment level set of γ just consists of the orbit together with
its stable/unstable manifolds. Then it is proved in [TZ2] that the mass of ϕµ in the shrinking
tube of radius hδ around γ with δ < 21 is ' (1−2δ) (see also [NV] for a closely related result in
two dimensions). Thus, the mass profile of such scarring integrable eigenfunctions only differs
by the numerical factor (1 − 2δ) from the mass profile of Gaussian beams. The difference
is that the ‘tails’ in the hyperbolic case are longer. Also the peak is logarithmically smaller
than in the elliptic case (a somewhat weaker statement is proved in [TZ2]).
Let us state the result precisely and briefly sketch the argument. It makes an interesting
comparison to the situation discussed later on of possible scarring in the Anosov case.
We denote by π : S ∗ M → M the standard projection and let π(γ) be the image of γ in
M . We denote by T² (π(Λ)) the tube of radius ² around π(Λ). For 0 < δ < 1/2, we introduce
a cutoff χδ1 (x; ~) ∈ C0∞ (M ) with 0 ≤ χδ1 ≤ 1, satisfying
• (i) supp χδ1 ⊂ T~δ (π(γ))
• (ii) χδ1 = 1 on T3/4~δ (π(γ)).
Theorem 5.2. Let γ be a hyperbolic closed orbit in (M, g) with quantum integrable ∆g , and
let {ϕµ } be an L2 normalized γ-sequence of joint eigenfunctions Then for any 0 ≤ δ < 1/2,
lim~→0 (Op~ (χδ1 )ϕµ , ϕµ ) ≥ (1 − 2δ).
5.1.1. Outline of proof. For simplicity we assume dim M = 2. Let χδ2 (x, ξ; ~) ∈ C0∞ (T ∗ M ; [0, 1])
be a second cutoff supported in a radius ~δ tube, Ω(~), around γ with Ω(~) ⊂ suppχδ1 and
such that χδ1 = 1 on suppχδ2 . Thus, χδ1 (x, ξ) ≥ χδ2 (x, ξ), for any (x, ξ) ∈ T ∗ M . By the
Garding inequality, there exists a constant C1 > 0 such that:
(59)
(Op~ (χδ1 )ϕµ , ϕµ) ) ≥ (Op~ (χδ2 )ϕµ , ϕµ ) − C1 ~1−2δ .
We now conjugate the right side to the model setting of S 1 ×R1 , i.e. the normal bundle Nγ
to γ. The conjugation is done by ~ Fourier integral operators and is known as conjugation
RECENT DEVELOPMENTS IN MATHEMATICAL QUANTUM CHAOS
31
to quantum Birkhoff normal form. In the model space, the conjugate of ∆g is a function
∂
of Ds = i∂s
along S 1 and the dilation operator Iˆh := ~(Dy y + yDy ) along R. By Egorov’s
theorem
(60)
(Op~ (χδ2 )ϕµ , ϕµ ) = |c(~)|2 (Op~ (χδ2 ◦ κ)uµ , uµ ) − C3 ~1−2δ
where uµ (y, s; ~) is the model joint eigenfunction of Ds , Iˆh , and c(~) is a normalizing constant.
This reduces our problem to estimating the explicit matrix elements (Op~ (χδ2 ◦κ)uµ , uµ ) of the
special eigenfunctions in the model setting. The operator Iˆh has a continuous spectrum with
generalized eigenfunctions y −1/2+iλ/~ . The eigenfunctions on the ‘singular’ level γ correspond
to λ ∼ E~. A caluclation shows that the mass in the model setting is given by
ÃZ
¯2 !
¯Z ∞
∞
¯ dξ
¯
1
(61)
Mh =
e−ix x−1/2+iλ/~ χ(x/~δ ξ)dx¯¯
χ(~ξ/~δ ) ¯¯
.
log ~
ξ
0
0
Classical analysis shows that the right side tends to 1 − 2δ as ~ → 0 if λ ∼ E~.
5.2. Comparison to Anosov case. This large mass profile may be a special feature of
integrable systems, reflecting the fact that the stable and unstable manifolds of the hyperbolic
closed orbit coincide. A heuristic picture of the mass concentration in this case is as follows:
Since the eigenfunction is a stationary state, its mass must be asymptotically invariant under
the geodesic flow. Since the flow compresses things exponentially in the stable direction and
expands things in the unstable direction, the mass can only concentrate on the fixed closed
geodesic and on the unstable manifold W u . But W u returns to γ as the stable manifold
W u (like a figure 8). Hence, the only invariant measure is the one supported on the closed
geodesic and the the mass can only concentrate there. Although it does not seem to have
been proved in detail yet, it is very plausible that the mass concentration in the integrable
case provides an upper bound for any (M, g), i.e. it has ‘extremal’ mass concentration.
In the Anosov case, the stable and unstable manifolds of γ are transverse, so the dynamical
picture is completely different. First, there is no obvious mechanism as in the integrable case
forcing mass of any sequence of eigenfunctions to concentrate on the Lagrangian manifold
formed by γ and W u in the Anosov case. If mass did concentrate around γ, it would still be
forced to concentrate on γ and on W u , but W u becomes dense in S ∗ M . Hence some of the
mass must spread out uniformly over S ∗ M and is lost from a neighborhood of γ. This makes
it plausible that one does not get mass concentration for eigenfunctions around hyperbolic
closed orbits of Anosov systems.
Yet, in the “cat map” analogue, there do exist scarring eigenfunctions [FNB, FN] for a
special sparse sequence of Planck constants. The multiplicities of the eigenvalues of the
cat map for this sequence saturate the bound ~−1 /| log ~|, and therefore one can build up
eigenfunctions with very unexpected properties. There is a surprising quantum mechanism
forcing producing concentration of special modes at hyperbolic fixed points which was discovered by Faure-Nonnenmacher-de Bièvre. The eigenfunction amplitude spreads out along
~−1/2 segments of W u . These segments are ~1/2 dense and so there is interference between
the amplitudes on close pieces of W u . The interference is constructive along W s and the
mass builds up there and then as in the integrable case gets recycled back to the hyperbolic
orbit.
32
STEVE ZELDITCH
It is not known whether this phenomenon occurs in the Riemannian setting. It is presumably related to the existence of sparse subsequences of eigenvalues with the same large
multiplicities.
6. Boundary quantum ergodicity and quantum ergodic restriction
In this section, we briefly discuss quantum ergodic restriction theorems. The general
question is whether restrictions of quantum ergodic eigenfunctions to hypersurfaces (or microlocally, to cross-sections of the geodesic flow) are quantum ergodic on the hypersurfaces.
We only consider here the case where the hypersurface is the boundary of a domain with
boundary, which was studied in [GL, HZ, Bu]. More general quantum ergodic restriction
theorems are given in [TZ3]. The boundary results play an important role in the recent
scarring results of A. Hassell for eigenfunctions on the stadium.
We thus consider the boundary values ubj of interior eigenfunctions

 ∆ uj = λ2j uj in Ω, huj , uk iL2 (Ω) = δjk ,

Buj |Y = 0,
Y = ∂Ω
of the Euclidean Laplacian ∆ on a compact piecewise smooth domain Ω ⊂ Rn and with
classically ergodic billiard map β : B ∗ Y → B ∗ Y , where Y = ∂Ω. Here Ah is a zeroth order
semiclassical pseudodifferential operator on Y . The relevant notion of boundary values
(i.e. Cauchy data) ubj depends on the boundary condition. We only consider the boundary
conditions

Dirichlet
 u|Y ,
(62)
Bu =

∂ν u|Y, Neumann
Let ∆B denote the positive Laplacian on Ω with boundary conditions Bu = 0. Then ∆B
has discrete spectrum 0 < λ1 < λ2 ≤ · · · → ∞, where we repeat each eigenvalue according
to its multiplicity, and for each λj we may choose an L2 normalized eigenfunction uj .
The algebra of observables in the boundary setting is the algebra Ψ0h (Y ) of zeroth order
semiclassical pseudodifferential operators on Y, depending on the parameter h ∈ [0, h0 ]. We
denote the symbol of A = Ah ∈ Ψ0h (Y ) by a = a(y, η, h). Thus a(y, η) = a(y, η, 0) is a
smooth function on T ∗ Y .
To each boundary condition B corresponds
• A specific notion of boundary value ubj of the eigenfunctions uj . We denote the
L2 -normalized boundary values by ûbj = ubj /||ubj ||.
• A specific measure dµB on B ∗ (Y ).
• A specific state ωB on the space Ψ0h (Y ) of semiclassical pseuodifferential operators of
order zero defined by
Z
4
ωB (A) =
a(y, η)dµB .
(63)
vol(S n−1 )vol(Ω) B ∗ Y
Here is a table of the relevant boundary value notions. In the table, dσ is the natural
symplectic volume measure on B ∗ Y . We also define the function γ(q) on B ∗ Y by
p
(64)
γ(q) = 1 − |η|2 , q = (y, η).
RECENT DEVELOPMENTS IN MATHEMATICAL QUANTUM CHAOS
33
Boundary Values
B
Bu
ub
dµB
−1
Dirichlet u|Y λ ∂ν u|Y γ(q)dσ
Neumann ∂ν u|Y
u|Y
γ(q)−1 dσ
The limit states are determined by dictated by the local Weyl law for the boundary
condition B.
Lemma 6.1. Let Ah be either the identity operator on Y or a zeroth order semiclassical
operator on Y with kernel supported away from the singular set. Then for any of the above
boundary conditions B, we have:
1 X
lim
hAhj ubj , ubj i = ωB (A), B = Neumann,
λ→∞ N (λ)
λj ≤λ
(65)
1 X
lim
hAhj ubj , ubj i = ωB (A), B = Dirichlet.
λ→∞ N (λ)
λ ≤λ
j
The boundary quantum ergodic theorem is:
Theorem 6.2. [HZ, GL, Bu] Let Ω ⊂ Rn be a bounded piecewise smooth manifold with
ergodic billiard map. Let {ubj } be the boundary values of the eigenfunctions {uj } of ∆B on
L2 (Ω) in the sense of the table above. Let Ah be a semiclassical operator of order zero on Y .
Then there is a subset S of the positive integers, of density one, such that
(66)
lim hAhj ubj , ubj i = ωB (A),
B = Neumann,
lim hAhj ubj , ubj i = ωB (A),
B = Dirichlet,
j→∞,j∈S
j→∞,j∈S
where hj = λ−1
and ωB is as in (63).
j
In the case A = I and for the Neumann boundary condition, we have
2vol(Y )
lim kubj k2L2 (Y ) =
,
j→∞,j∈S
vol(Ω)
while for the Dirichlet boundary condition,
2vol(Y )
lim kubj k2L2 (Y ) =
.
j→∞,j∈S
nvol(Ω)
6.1. Sketch of proof. The fact that the quantum limit state ωB and the corresponding
measure dµB are not the natural symplectic volume measure dσ on B ∗ Y is due to the fact
that the quantum dynamics is defined by an endomorphism rather than an automorphism of
the observable algebra. In the Neumann case, the dynamics are generated by the operator
Fh on Y with kernel
∂
G0 (y, y 0 , h−1 ), y 6= y 0 ∈ Y, where
Fh (y, y 0 ) = 2
∂νy
(67)
i
(1)
G0 (y, y 0 , λ) = λd−2 (2πλ|z − z 0 |)−(d−2)/2 Had/2−1 (λ|z − z 0 |)
4
34
STEVE ZELDITCH
is the free outgoing Green function on Rd . These are (almost) semi-classical Fourier integral
operators whose phase functions, the boundary distance function db (y, y 0 ) = |y−y 0 (y, y 0 ∈ Y )
generates the billiard map. It is well known that this operator leaves the boundary values
of Neumann eigenfunctions ubj invariant:
Fhj ubj = ubj ,
(68)
j = 1, 2, . . .
It follows that the states
(69)
ρj (A) := hAhj ubj , ubj i
are invariant for Fhj . The family {Fh } defines a semiclassical Fourier integral operator
associated to the billiard map β (for convex Ω), plus some negligible terms. The quantum
dynamics on Ψ0h (Y ) is thus generated by the conjugation
(70)
αhj (Ahj ) = Fh∗j Ahj Fhj
This is analogous to the interior dynamics generated by the wave group UBt with the boundary
conditions, but unlike UBt , Fh is not unitary or even normal. Indeed, the zeroth order part
of Fh∗ Fh is a semiclassical pseudodifferential operator with a non-constant symbol.
The Egorov type result for the operator Fh is as follows: Let β denote the billiard map
on B ∗ Y o and let Ah = Op(ah ) be a zeroth order operator whose symbol a(y, η, 0). Let γ be
given by (64). Then
Fh∗ Ah Fh = Ãh + Sh ,
where Ãh is a zeroth order pseudodifferential operator and kSh kL2 →L2 ≤ Ch. The symbol of
Ãh is
(
γ(q)γ(β(q))−1 a(β(q)), q ∈ B ∗ Y
(71)
ã =
0,
q∈
/ B ∗ Y.
This is a rigorous version of the statement that Fh quantizes the billiard ball map. This
Egorov theorem is relevant to the Neumann boundary problem. In the Dirichlet case, the
relevant operator is Fh∗ .
The unusual transformation law of the symbol reflects the fact that (70) is not an automorphism. In the spectral theory of dynamical systems, one studies the dynamics of the
billiard map β on B ∗ Y through the associated ‘Koopman’ operator
U : L2 (B ∗ Y, dσ) → L2 (B ∗ Y, dσ), Uf (ζ) = f (β(ζ)).
Here, dσ denotes the usual β-invariant symplectic volume measure on B ∗ Y . From the
symplectic invariance it follows that U is a unitary operator. When β is ergodic, the unique
invariant L2 -normalized eigenfunction is a constant c.
However, Egorov’s theorem (71) in the boundary reduction involves the positive function
γ ∈ C ∞ (B ∗ Y ), and the relevant Koopman operator is
T f (ζ) =
γ(ζ)
f (β(ζ)).
(γ(β(ζ)))
Then T is not unitary on L2 (B ∗ Y, dσ). Instead one has:
• (i) The unique positive T -invariant L1 function is given by γ.
RECENT DEVELOPMENTS IN MATHEMATICAL QUANTUM CHAOS
35
• (ii) T is unitary relative to the inner product hh, ii on B ∗ Y defined by the measure
dν = γ −2 dσ.
• (iii) When β is ergodic, the orthogonal projection P onto the invariant L2 -eigenvectors
has the form
Z
hhf, γii
1
P (f ) =
γ=
[
f γ −1 dσ] γ = cωNeu (f ) γ
hhγ, γii
vol(B ∗ Y ) B ∗ Y
where c is as in (112).
The proof of Theorem 6.2 then runs along similar lines to that of Theorem 4.1 but adjusted
to the fact that that T is not unitary.
7. Hassell’s scarring result for stadia
This section is an exposition of Hassell’s scarring result for the Bunimovich stadium. We
follow [Has] and [Z7].
A stadium is a domain X = R ∪ W ⊂ R2 which is formed by a rectangle R = [−α, α]x ×
[−β, β]y and where W = W−β ∪ Wβ are half-discs of radius β attached at either end. We fix
the height β = π/2 and let α = tβ with t ∈ [1, 2]. The resulting stadium is denoted Xt .
It has long been suspected that there exist exceptional sequences of eigenfunctions of X
which have a singular concentration on the set of “bouncing ball” orbits of R. These are the
vertical orbits in the central rectangle that repeatedly bounce orthogonally against the flat
part of the boundary. The unit tangent vectors to the orbits define an invariant Lagrangian
submanifold with boundary Λ ⊂ S ∗ X. It is easy to construct approximate eigenfunctions
which concentrate microlocally on this Lagrangian submanifold. Namely, let χ(x) be a
smooth cutoff supported in the central rectangle and form vn = χ(x) sin ny. Then for any
pseudo-differential operator A properly supported in X,
Z
hAvn , vn i →
σA χdν
Λ
where dν is the unique normalized invariant measure on Λ.
Numerical studies suggested that there also existed genuine eigenfunctions with the same
limit. Recently, A. Hassell has proved this to be correct for almost all stadia.
Theorem 7.1. The Laplacian on Xt is not QUE for almost every t ∈ [1, 2].
We now sketch the proof and develop related ideas on quantum unique ergodicty. The
main idea is that the existence of the scarring bouncing ball quasi-modes implies that either
• There exist actual modes with a similar scarring property, or
• The spectrum has exceptional clustering around the bouncing ball quasi-eigenvalues
n2 .
Hassell then proves that the second alternative cannot occur for most stadia. We now
explain the ideas in more detail.
We first recall that a quasi-mode {ψk } is a sequence of L2 -normalized functions which
solve
||(∆ − µ2k )ψk ||L2 = O(1),
36
STEVE ZELDITCH
for a sequence of quasi-eigenvalues µ2k . By the spectral theorem it follows that there must
exist true eigenvalues in the interval [µ2k − K, µ2k + K] for some K > 0. Moreover, if Ẽk,K
denotes the spectral projection for ∆ corresponding to this interval, then
||Ẽk,K ψk − ψk ||L2 = O(K −1 ).
To maintain consistency with (6), i.e. with our use √
of frequencies µk rather
µ2k ,
p than energies
p
we re-phrase this in terms of the projection Ek,K for ∆ in the interval [ µ2k − K, µ2k + K].
For fixed K, this latter interval has width µKk .
Given a quasimode {ψk }, the question arises of how many true eigenfunctions it takes to
build the quasi-mode up to a small error.
Definition: We say that a quasimode {ψk } of order 0 with ||ψk ||L2 = 1 has n(k) essential
frequencies if
(72)
ψk =
n(k)
X
ckj ϕj + ηk , ||ηk ||L2 = o(1).
j=1
To be a quasi-mode of order zero, the frequencies λj of the ϕj must come from an interval
[µk − µKk , µk + µKk ]. Hence the number of essential frequencies is bounded above by the
√
number n(k) ≤ N (k, Kk ) of eigenvalues in the interval. Weyl’s law for ∆ allows considerable
clustering and only gives N (k, Kk ) = o(k) in the case where periodic orbits have measure zero.
For instance, the quasi-eigenvalue might be a true eigenvalue with multiplicity saturating
the Weyl bound. But a typical interval has a uniformly bounded number of ∆-eigenvalues
in dimension 2 or equivalently a frequency interval of with O( µ1k ) has a uniformly bounded
number of frequencies. The dichotomy above reflects the dichotomy as to whether exceptional
clustering of eigenvalues occurs around the quasi-eigenvalues n2 of ∆ or whether there is a
uniform bound on N (k, δ).
Proposition 7.2. If there exists a quasi-mode {ψk } of order 0 for ∆ with the properties:
• (i) n(k) ≤ C, ∀ k;R
• (ii) hAψk , ψk i → S ∗ M σA dµ where dµ 6= dµL .
Then ∆ is not QUE.
The proof is based on the following lemma pertaining to near off-diagonal Wigner distributions. It gives an “everywhere” version of the off-diagonal part of Theorem 4.1 (2).
Lemma 7.3. Suppose that g t is ergodic
√ and ∆ is QUE. Suppose that {(λir , λjr ), ir 6= jr } is a
sequence of pairs of eigenvalues of ∆ such that λir − λjr → 0 as r → ∞. Then dWir ,jr → 0.
Proof. Let {λi , λj } be any sequence of pairs with the gap λi − λj → 0. Then by Egorov’s
theorem, any weak* limit dν of the sequence {dWi,j } is a measure invariant under the geodesic
flow. The weak limit is defined by the property that
Z
∗
(73)
hA Aϕi , ϕj i →
|σA |2 dν.
S∗M
If the eigenfunctions are real, then dν is a real (signed) measure.
RECENT DEVELOPMENTS IN MATHEMATICAL QUANTUM CHAOS
37
We now observe that any such weak* limit must be a constant multiple of Liouville measure
dµL . Indeed, we first have:
(74)
|hA∗ Aϕi , ϕj i| ≤ |hA∗ Aϕi , ϕi i|1/2 |hA∗ Aϕj , ϕj i|1/2 .
Taking the limit along the sequence of pairs, we obtain
Z
Z
2
(75)
|
|σA | dν| ≤
|σA |2 dµL .
S∗M
S∗M
It follows that dν << dµL (absolutely continuous). But dµL is an ergodic measure, so if
dν = f dµL is an invariant measure with f ∈ L1 (dµL ), then f is constant. Thus,
(76)
dν = CdµL , for some constant C.
We now observe that C = 0 if ϕi ⊥ϕj (i.e. if i 6= j). This follows if we substitute A = I in
(73), use orthogonality and (76).
¤
We now complete the proof of the Proposition by arguing by contradiction. The frequencies
must come from a shrinking frequency interval, so the hypothesis of the Proposition is
satisfied. If ∆ were QUE, we would have (in the notation of (72):
P
hAψk , ψk i = n(k)
i,j=1 ckj cki hAϕi , ϕj i + o(1)
=
=
Pn(k)
j=1
R
S∗M
c2kj hAϕj , ϕj i +
Pn(k)
i6=j=1 ckj cki hAϕi , ϕj i
+ o(1)
σA dµL + o(1),
by Proposition 7.3. This contradicts (ii). In the last line we used
since ||ψk ||L2 = 1.
Pn(k)
j=1
|ckj |2 = 1 + o(1),
QED
7.1. Proof of Hassell’s scarring result. We apply and develop this reasoning in the case
of the stadium. The quasi-eigenvalues of the Bunimovich stadium corresponding to bouncing
ball quasi-modes are n2 independently of the diameter t of the inner rectangle.
By the above, it suffices to show that that there exists a sequence nj → ∞ and a constant
M (independent of j) so that there exist ≤ M eigenvalues of ∆ in [n2j − K, n2j + K]. An
somewhat different argument is given in [Has]q
in this case: For each nj there exists a normalized eigenfunction ukj so that hukj , vkj i ≥ 34 M . It suffices to choose the eigenfunction
with eigenvalue in the interval with the largest component in the direction of vkj . There
exists one since
3
||Ẽ[n2 −K,n2 +K vn || ≥ .
4
The sequence {unk } cannot be Liouville distributed. Indeed, for any ² > 0, let A be a
self-adjoint semi-classical pseudo-differential operator properly supported in the rectangle so
38
STEVE ZELDITCH
that σA ≤ 1 and so that ||(Id − A)vn || ≤ ² for large enough n. Then
¯
¯2
hA2 ukj , ukj i = ||Aukj ||2 ≥ ¯hAukj , vkj i¯
´2
¯
¯2 ¡
¢2 ³q 3
= ¯hukj , Avkj i¯ ≥ |hukj , vkj i| − ² ≥
M
−
²
.
4
Choose a sequence of operators A such that ||(Id − A)vn || → 0 and so that the support of
σA shrinks to the set of bouncing ball covectors. Then the mass of any quantum limit of
{unk } must have mass ≥ 43 M on Λ.
Thus, the main point is to eliminate the possibility of exceptional clustering of eigenvalues
around the quasi-eigenvalues. In fact, no reason is known why no exceptional clustering
should occur. Hassell’s idea is that it can however only occur for a measure zero set of
diameters of the inner rectangle. The proof is based on Hadamard’s variational formula for
the variation of Dirichlet or Neumann eigenvalues under a variation of a domain. In the
case at hand, the stadium is varied by horizontally (but not vertically) expanding the inner
rectangle. In the simplest case of Dirichlet boundary conditions, the eigenvalues are forced
to decrease as the rectangle is expanded. The QUE hypothesis forces them to decrease at
a uniform rate. But then they can only rarely cluster at the fixed quasi-eigenvalues n2 . If
this ever happened, the cluster would move left of n2 and there would not be time for a new
cluster to arrive.
Here is a more detailed sketch. Under the variation of Xt with infinitesimal variation
vector field ρt , Hadamard’s variational formula gives,
Z
dEj (t)
=
ρt (s)(∂n uj (t)(s))2 ds.
dt
∂Xt
Then
Ej−1
d
Ej (t) = −
dt
Z
∂Xt
ρt (s)ubj (s)2 ds.
j
Let A(t) be the area of St . By Weyl’s law, Ej (t) ∼ c A(t)
. Since the area of Xt grows
E
j
. Theorem 6.2 gives the asymptotics individually
linearly, we have on average Ėj ∼ −C A(t)
for almost all eigenvalues. Let
Z
fj (t) =
ρt (s)|ubj (t; s)|2 ds.
∂Xt
1
Then Ėj = −Ej fj . Then Theorem 6.2 implies that |ubj |2 → A(t)
weakly on the boundary
along a subsequence of density one. QUE is the hypothesis that this occurs for the entire
sequence, i.e.
Z
k
> 0, k :=
ρt (s)ds.
fj (t) →
A(t)
∂St
Hence,
Ėj
= −kA(t)(1 + o(1)), j → ∞.
E
Hence there is a lower bound to the velocity with which eigenvalues decrease as A(t) increases.
Eigenvalues can therefore not concentrate in the fixed quasi-mode intervals [n2 − K, n2 + K]
for all t. But then there are only a bounded number of eigenvalues in this interval; so
RECENT DEVELOPMENTS IN MATHEMATICAL QUANTUM CHAOS
39
Proposition 7.2 implies QUE for the other Xt . A more detailed analysis shows that QUE
holds for almost all t.
8. Matrix elements of Fourier integral operators
Our main focus in this survey goal is on the limits of diagonal matrix elements (12)
of pseudodifferential operators A or order zero relative to an orthonormal basis {ϕj } of
eigenfunctions. Difficult as it is, the limiting behavior of such matrix elements is one of
the more accessible properties of eigenfunctions. To obtain more information, it would be
useful to expand the class of operators A for which one can study matrix elements. In §5,
we were essentially expanding the class from classical polyhomogeneous pseudo-differential
operators to those in the “small-scale” calculus, i.e. supported in hδ tubes around closed
geodesics. Another way to expand the class of A is to consider matrix elements of Fourier
integral operators. Motivation to consider matrix elements of Fourier integral operators
comes partly from the fact that Hecke operators are Fourier integral operators, and this
section is a preparation for the next section on Hecke operators §9. Other examples arise
in quantum ergodic restriction theorems. Details on the claims to follow will be given in a
forthcoming paper.
By an FIO (Fourier integral operator) we mean an operator whose Schwartz kernel of F
may be locally represented as a finite sum of oscillatory integrals,
Z
KF (x, y) ∼
eiϕ(x,y,θ) a(x, y, θ)dθ
RN
for some homogeneous phase ϕ and amplitude a. The only example we have seen so far in
this survey is the wave group U t and of course pseudo-differential operators.
The phase is said to generate the canonical relation
C = {(x, ϕ0x , y, −ϕ0y ) : ϕ0ξ (x, y, θ) = 0} ⊂ T ∗ M × T ∗ M.
The oscillatory integral also determines the principal σF of F , a 1/2 density along C. The
data (C, σF ) determines F up to a compact operator. In the case of the wave group, C is
the graph of g t and the symbol is the canonical volume half-density on the graph. In the
case of ψDO’s, the canonical relation is the diagonal (i.e. the graph of the identity map)
and the symbol is the one defined above. We denote by I 0 (M, ×M, C) the class of Fourier
integral operators of order zero and canonical relation C.
Hecke operators are also FIO’s of a simple kind. As discussed in the next section, in that
case C is a local canonical graph, i.e. the projections
πX : C → T ∗ M, πY : C → T ∗ M
are branched covering maps. Equivalently, C is the graph of a finitely many to one correspondence χ : T ∗ M → T ∗ M . Simpler examples of the same type include (i) F = Tg is
translation by an isometry of a Riemannian manifold (M, g) possessing an isometry, or (ii)
P
T f (x) = kj=1 f (gj x)+f (gj−1 x) on S n corresponding to a finite set {g1 , . . . , gk } of isometries.
The latter has been studied by Lubotzky-Phillips-Sarnak.
40
STEVE ZELDITCH
8.1. Matrix elements as linear functionals. The matrix elements
(77)
ρj (F ) := hF ϕj , ϕj i
define continuous linear functionals
ρj : I 0 (M × M, C) → C
(78)
with respect to the operator norm. It is simple to see that any limit ρ∞ of the sequence of
functionals ρj are functionals only of the principal symbol data (C, σF ) which is bounded by
the supremum of σF . Thus, as in the pseudo-differential case, ρ∞ is a measure on the space
1/2
ΩC of continuous half-densities on C.
√
The limiting behavior of ρj (F ) depends greatly on whether [F, ∆] = 0 (or is of negative
order) or whether it has the same order as F . In effect, this is the issue of whether the
canonical relation C underlying F is invariant under the geodesic flow or not. For instance,
if C is the graph of a canonical transformation χ and if χ moves the energy surface Sg∗ M =
{(x, ξ) : |ξ|g = 1} to a new surface χ(Sg∗ M ) disjoint from Sg∗ M then hF ϕj , ϕj i → 0. On
√
the other√ hand, hAeit ∆ ϕj , ϕj i = eitλj hAϕj , ϕj i so one gets the same quantum limits for
F = Aeit ∆ as for A but must pick sparser subsequences to get the limit.
Hecke operators have the property that [F, ∆] = 0. This implies that the canonical relation
C is invariant under g t . In general this is the only condition we need to study weak limits. The
eigenfunction linear functionals ρk are invariant in the the sense that ρk (U −t F U t ) = ρk (F ).
Proposition 8.1. Let C ⊂ T ∗ M × T ∗ M be a local canonical graph, equipped with the
pull-back of the symplectic volume measure on T ∗ M . Assume C is invariant under g t . Let
F ∈ I 0 (M × M, C) and identify its symbol with a scalar function relative to the graph halfdensity. Let ρ∞ be a weak* limit of the functionals ρj on I 0 (M × N, C). Then there exists
a complex measure ν on SC = C ∩ S ∗ M × S ∗ M of mass ≤ 1 such that
Z
ρ∞ (F ) =
σF dν,
SC
which satisfies
g∗t dν
= dν.
The proof is similar to that in the ψDO case.
We note that I 0 (M × M, C) is a right and left module over Ψ0 (M ). Given F ∈ I 0 (M ×
M, C) we consider the operators AF, F A ∈ I 0 (M × M, C) where A ∈ Ψ0 (X). We obtain a
useful improvement on Proposition 8.1.
Proposition 8.2. With the same hypotheses as in Proposition 8.1. Assume further that F
is self-adjoint and that ρj (AF ) ∼ ρj (F A). Then we have:
∗
aσ) = ρ∞ (πY∗ aσ).
ρ∞ (πX
This holds since
∗
ρ∞ (πX
aσ) ∼ ρj (AF ) = hAF ϕj , ϕj i = hF ∗ A∗ ϕj , ϕj i = hF A∗ ϕj , ϕj i ∼ πY∗ aσ
since σA∗ ∼ σA and since σF = σF .
Let us consider some simple examples. First, suppose that F = Tg + Tg∗ where g is an
isometry. Then C is the union of the graph of the lift of g to T ∗ M (by its derivative)
and the inverse graph. The above Proposition then says that g∗ ν = ν. This is obvious by
RECENT DEVELOPMENTS IN MATHEMATICAL QUANTUM CHAOS
41
Egorov’s theorem, and we can regard the Proposition as a generalization of Egorov’s theorem
to symplectic correspondences. As a second example, let F = U t . Then hAF ϕj , ϕj i =
eitλj hAϕj , ϕj i, so we see that sequences with unique quantum limits are sparser than in the
pseudo-differential case.
9. QUE of Hecke eigenfunctions
Our next topic is Hecke eigenfunctions. A detailed treatment would have to involve adelic
dynamics, higher rank measure rigidity in the presence of positive entropy [LIND, LIND2]
and L-functions [HS, Sound1]. Both areas are outside the scope of this survey. Fortunately,
Lindenstrauss has written an expository article on adelic dynamics and higher rank rigidity for an earlier Current Developments in Mathematics [LIND2], and Soundararajan has
recently lectured on his L-function results in a 2009 Clay lecture. Sarnak has recently written exposition of the new results of Soundararajan and Holowinsky-Soundararajan results is
[Sar3].
So we continue in the same vein of explaining the microlocal (phase space) features of the
Hecke eigenvalue problem. The distinguishing feature of Hecke eigenfunctions is that they
have a special type of symmetry. As a result, their quantum limit measures have a special
type of symmetry additional to geodesic flow invariance. The analysis of these additional
symmetries is a key input into Lindenstrauss’ QUE result. The main purpose of this section is
to derive the exact symmetry. In fact, it appears that the exact symmetry was not previously
determined; only a quasi-invariance condition of [RS] has been employed.
We begin by recalling the definition of a Hecke operator. Let Γ be a co-compact or cofinite
discrete subgroup of G = P SL(2, R) and let XΓ be the corresponding compact (or finite area)
hyperbolic surface. An element g ∈ G, g ∈
/ Γ is said to be in the commensurator Comm(Γ)
if
Γ0 (g) := Γ ∩ g −1 Γg
is of finite index in Γ and g −1 Γg. More precisely,
Γ=
d
[
Γ0 (g)γj , (disjoint),
j=1
or equivalently
ΓgΓ =
d
[
Γαj , where αj = gγj .
j=1
It is also possible to choose αj so that ΓgΓ =
(non-Galois) covers:
Sd
j=1
αj Γ. We then have a diagram of finite
Γ0 (g)\H
π.
(79)
Γ\H
&ρ
⇐⇒
g −1 Γg\H.
Here,
π(Γ0 (g)z) = Γz, ρ(Γ0 (g)z) = gΓg −1 (gγj z) = gΓz,
42
STEVE ZELDITCH
where in the definition of ρ any of the γj could be used. The horizontal map is z → g −1 z.
A Hecke operator
(80)
Tg f (x) =
d
X
f (gγj x) : L2 (XΓ ) → L2 (XΓ )
j=1
is the Radon transform ρ∗ π ∗ of the diagram. We note as a result that
Tg−1 u(z) = u(g −1 z)
takes Γ-invariant functions to g −1 Γg-invariant functions. The Hecke operator is thus an
averaging operator over orbits of the Hecke correspondence, which is the multi-valued holomorphic map
Cg (z) = {α1 z, . . . , αd z}.
Its graph
Γg = {(z, αj z) : z ∈ XΓ )}
is an algebraic curve in XΓ × g −1 Xg.
The Hecke operators Tg form a commutative ring as g varies over Comm(Γ). By a Hecke
eigenfunction is meant a joint eigenfunction of ∆ and the ring of Hecke operators:
Tg uj = µj (g)uj , ∆uj = −λ2j uj .
9.1. Quantum limits on Γ0 (g)\H. We now considered symmetries of quantum limits of
Hecke eigenfunctions. We assume uj is a Hecke eigenfunction. It is convenient to modify the
definitions of the functionals ρk as follows:
 0
hAπ ∗ uj ,ρ∗ uj i
ρ
(A)
=
=

j
hπ ∗ uj ,ρ∗ uj i





hAπ ∗ u ,π ∗ u i
σj0 (A) = hπ∗ ujj,ρ∗ uj ji ,





 0
hAπ ∗ u ,ρ∗ u i
τj (A) = hρ∗ uj j,ρ∗ uj ji
(81)
hρ∗ Aπ ∗ uj ,uj i
,
hπ ∗ uj ,ρ∗ uj i
on A ∈ Ψ0 (Γ0 (g)\H). Thus, if A0 ∈ Ψ0 (XΓ ) we have,
hA0 ρ∗ π ∗ uj , uj i
= hA0 uj , uj i.
hρ∗ π ∗ uj , uj i
(82)
Since Hecke operators are FIO’s, limit measures of hATg uj , uj i and hTg Auj , uj i live on the
canonical relation underlying Tg , which is the lift of the graph of the Hecke correspondence
to T ∗ XΓ . It is convenient to restate the result by forming a second diagram
Γg ⊂ XΓ × g −1 Γg\H
π1 .
(83)
Γ\H
& ρ1
⇐⇒
g −1 Γg\H.
RECENT DEVELOPMENTS IN MATHEMATICAL QUANTUM CHAOS
43
Here, π1 , ρ1 are the natural projections. In the case of arithmetic groups such as SL(2, Z),
Comm(Γ) is dense in G, i.e. there are many Hecke operators. The map
ι : Γ0 (g)\H → Γg
defined by
ι(z) = (π(z), ρ(z)),
is a local diffeomorphic parametrization, and π1 ι = π, ρ1 ι = ρ.
The general Proposition 8.2 then specializes to:
Proposition 9.1. If ω 0 is a weak* limit of ρ0j , and if ω is the weak* limit on XΓ then
π∗ ω 0 = ρ∗ ω 0 = ω.
We may reformulate the proposition as follows: if we lift measures and functions to the
universal cover H, then the data of the Hecke limit measures are a finite set of real signed
measures {dνk } on the different copies of S ∗ H. The latter must project to the former under
both projections, so we get the exact invariance property,
(84)
ν=
d
X
ναj =
j=1
d
X
αj∗ ναj ,
j=1
for some complex measures νj on the j sheet of the cover. The sum of the masses of the
νj must add up to one. This is simpler to see for Hecke operators on S 2 . There we also
have some number r of copies of S 2 and r isometries gj and define the Hecke operator (of
Lubotzky-Phillips-Sarnak) as in (80). Then a quantum limit measure ν0 on S ∗ S 2 is the
projection of r limit measures on the copies of S ∗ S 2 under both projections.
Previously, only a quasi-invariance property of quantum limit measures of Hecke eigenfunctions was proved by Rudnick-Sarnak [RS], and used in [BL, W3] and elsewhere. In the
notation above, it may be stated as
X
(85)
ν(E) ≤ d
|αj∗ ναj (E)|.
j
As a simple application of (84) (which was also clear from the quasi-invariance condition),
we sketch a proof that a quantum limit measure ν of a sequence of Hecke eigenfunctions
P for
Γ = SL(2, Z) cannot be a periodic orbit measure µγ . If it were, it would have to be
νj as
above. Each νj is an invariant
signed
measure
so
by
ergodic
decomposition,
each
must
P
P have
the form cj µγ + τj where
cj = 1, where τj are singular with respect to µγ and j τj = 0.
But also the images of νj under the αj must have the same property. This is only possible
if αj (γ) = γ up to a translation by β ∈ Γ whenever νj 6= 0. At least one of the cj 6= 0. At
this point, one must actually consider the elements αj in the Hecke operator. In the case
of P SL(2, Z) they are parabolic elements plus one elliptic element. But only a hyperbolic
element can fix a geodesic (its axis).
It would be interesting to see if one can determine how the ναj are related to each other.
Lindenstrauss’s QUE result (in the next section) implies that that (Tg )∗ ν = ν, which appears
to say that the ναj are all the same. An apriori proof of this would simplify the proof of
QUE.
44
STEVE ZELDITCH
9.2. QUE results of Lindenstrauss, Soundararajan (and Holowinsky-Soundararajan).
We now briefly recall the QUE results on Hecke eigenfunctions.
Theorem 9.2. (E. Lindenstrauss [LIND]) Let XΓ be a compact arithmetic hyperbolic surface. Then QUE holds for Hecke eigenfunctions, i.e. the entire sequence of Wigner distributions of Hecke eigenfunctions tends to Liouville measure.
In the non-compact finite area case of Γ = P SL(2, Z), Lindenstrauss proved:
Theorem 9.3. (E. Lindenstrauss [LIND]) Let XΓ be the arithmetic hyperbolic surface defined by a congruence subgroup. Then any weak* limit of the sequence of Wigner distributions
of Hecke eigenfunctions is a constant multiple of Liouville measure (the constant may depend
on the sequence).
Note that the quantum ergodicity theorem does not apply in this finite area case. In the
arithmetic case there is a discrete spectrum corresponding to cuspidal eigenfunctions and a
continuous spectrum corresponding to Eisenstein series. The following was proved in [Z8]:
Theorem 9.4. Let XΓ be the arithmetic hyperbolic surface defined by a congruence subgroup,
and let {ϕj } be any orthonormal basis of cuspidal eigenfunctions. Then for any pseudodifferential operator of order zero and with compactly supported symbol,
X
1
|hAϕj , ϕj i − ω(A)|2 → 0,
Nc (λ) j:λ ≤λ
j
where Nc (λ) ∼ |XΓ |λ2 is the number of cuspidal eigenvalues ≤ λ.
Recent work of Soundararajan that the constant equals one, i.e. there is no mass leakage
at infinity. Hence the theorem is improved to
Theorem 9.5. (E. Lindenstrauss [LIND], Soundarajan [Sound1]) Let XΓ be the arithmetic
hyperbolic surface defined by a congruence subgroup. Then any weak* limit of the sequence
of Wigner distributions of Hecke eigenfunctions is is normalized Liouville measure.
Holowinsky [Hol] and Holowinsky-Soundararajan [HS] have proven QUE results for holomorphic forms (i.e. for the discrete series; see §1.15.4). However, the methods are entirely
different; they are based on the theory of L-functions and Poincaré series. The Hecke property is exploited through multiplicativity of Fourier coefficients.
9.3. Hecke QUE and multiplicities. A natural question (raised at the Clay talk of
Soundararajan) is whether QUE holds for any orthonormal basis of eigenfunctions of an
arithmetic quotient if it holds for the Hecke eigenbasis and if the discrete spectrum of ∆ has
bounded multiplicities. The proof that this is true is a simple modification of Proposition
7.2 and we pause to sketch it now.
Proposition 9.6. Suppose that one orthonormal basis of ∆ is QUE, and that the eigenvalues have uniformly bounded multiplicities. Then all orthonormal bases are QUE.
Proof:
As in Proposition 7.2, consider sequences {(λir , λjr ), ir 6= jr } of pairs of eigenvalues
√
of ∆ with λir = λjr . We then have that dWir ,jr → 0.
This has implications for all orthonormal basis as long as eigenvalue multiplicities are
uniformly bounded. We run the argument regarding modes and quasi-modes but use the
RECENT DEVELOPMENTS IN MATHEMATICAL QUANTUM CHAOS
45
QUE orthonormal basis in the role of modes and any other orthonormal basis {ψk } in the
role of quasi-modes.
Lemma 9.7. If the eigenvalue multiplicities are uniformly bounded and there exists a sequence
of eigenfunctions {ψk }
• (i) n(k) ≤ C, ∀ k;R
• (ii) hAψk , ψk i → S ∗ M σA dµ where dµ 6= dµL .
Then no orthonormal basis of eigenfunctions of ∆ is QUE.
Proof. We argue by contradiction. If {ϕj } were QUE, we would have (in the notation of
(72):
P
hAψk , ψk i = n(k)
i,j=1 ckj cki hAϕi , ϕj i + o(1)
=
=
Pn(k)
j=1
R
S∗M
c2kj hAϕj , ϕj i +
Pn(k)
i6=j=1 ckj cki hAϕi , ϕj i
σA dµL + o(1),
by Proposition 9.6. This contradicts (ii). In the last line we used
since ||ψk ||L2 = 1.
+ o(1)
Pn(k)
j=1
|ckj |2 = 1 + o(1),
¤
10. Variance estimates: Rate of quantum ergodicity and mixing
A quantitative refinement of quantum ergodicity is to ask at what rate the sums in Theorem 4.1(1) tend to zero, i.e. to establish a rate of quantum ergodicity. More generally, we
consider ‘variances’ of matrix elements. For diagonal matrix elements, we define:
X
1
|hAϕj , ϕj ) − ω(A)|2 .
(86)
VA (λ) :=
N (λ) j:λ ≤λ
j
In the off-diagonal case one may view |hAϕi , ϕj i|2 as analogous to |hAϕj , ϕj ) − ω(A)|2 . However, the sums in (57) are double sums while those of (86) are single. One may also average
over the shorter intervals [λ, λ + 1].
10.1. Quantum chaos conjectures. It is implicitly conjectured by Feingold-Peres in [FP]
(11) that
E −E )
(87)
CA ( i ~ j )
|hAϕi , ϕj i| '
,
2πρ(E)
2
R∞
where CA (τ ) = −∞ e−iτ t hV t σA , σA idt. In our notation, λj = ~−1 Ej (with ~ = λ−1
and
j
2
Ej = λj ) and ρ(E)dE ∼ dN (λ).
On the basis of the analogy between |hAϕi , ϕj i|2 and |hAϕj , ϕj ) − ω(A)|2 , it is conjectured
in [FP] that
CA−ω(A)I (0)
.
VA (λ) ∼ n−1
λ vol(Ω)
The idea is that ϕ± = √12 (ϕi ± ϕj ) have the same matrix element asymptotics as eigenfunctions when λi − λj is sufficiently small. But then 2hAϕ+ , ϕ− i = hAϕi , ϕi i − hAϕj , ϕj i when
46
STEVE ZELDITCH
A∗ = A. Since we are taking a difference, we may replace each matrix element by hAϕi , ϕi i
by hAϕi , ϕi i − ω(A) (and also for ϕj ). The conjecture then assumes that hAϕi , ϕi i − ω(A)
has the same order of magnitude as hAϕi , ϕi i − hAϕj , ϕj i.
10.2. Rigorous results for Hecke eigenfunctions. Let XΓ be the arithmetic modular
surface, with Γ = SL(2, Z), and let Tn , n > 1 denote the family of Hecke operators. In addition to the Laplace eigenvalue problem (28), the Maass-Hecke eigenforms are eigenfunctions
of the Hecke operators,
Tn ϕ = λϕ (n)ϕ.
In the cusp, they have Fourier series expansions
√ X
ϕ(z) = y
ρϕ (n)Kir (2π|n|y)e(nx)
n6=0
where Kir is the K-Bessel function and ρϕ (n) = λϕ (n)ρϕ (1). A special case of the variance
sums (86) is the configuration space variance sum,
X Z
Sψ (λ) =
| ψϕ2 dV |2
λj ≤λ
Following the notation of [LS2] and [Zh],we put dµn = ϕ2rn dV .
∞
Let ψ ∈ C0,0
(S ∗ XΓ ) and consider the distribution of
Z T
1
√
ψ(g t (x, ξ))dt.
T 0
Ratner’s central limit theorem for geodesic flows implies that this random variable tends
to a Gaussian with mean 0 and the variance given by the non-negative Hermitian form on
∞
C0,0
(Y ):
µ µ t
¶¶
Z ∞Z
0
e2
VC (ϕ, ψ) =
ϕ g
ψ(g)dgdt.
t
0 e− 2
−∞ Γ\SL(2,R)
∞
Restrict VC to C0,0
(X). Different irreducible representations are orthogonal under the quadratic form. Maass cusp forms ϕ are eigenvectors of VC . If the Laplace eigenvalue is 14 + r2 ,
then the VC eigenvalue is
|Γ( 14 − ir2 )|4
VC (ϕ, ϕ) =
2π|Γ( 12 − ir)|2
In [LS, LS2], Luo and Sarnak studied the quantum variance for holomorhic Hecke eigenforms, i.e. holomorphic cusp forms in Sk (Γ) of even integral weight k for Γ. In this setting,
the weight of the cusp form plays the role of the Laplace eigenvalue. They proved that for
∞
(X), as the weight K → ∞,
ϕ, ψ ∈ C0,0
X X
L(1, Sym2 f )µf (ϕ)µf (ψ) ∼ B(ϕ, ψ)K,
k≤K f ∈Hk
∞
(X). B is diagonalized by the
where B(ϕ, ψ) is a non-negative Hermitian form on C0,0
orthonormal basis of Maass-Hecke cusp forms and the eigenvalues of B at ψj is π2 L( 12 , ψj ).
For the notation L(1, Sym2 ) we refer to [LS2, Zh].
RECENT DEVELOPMENTS IN MATHEMATICAL QUANTUM CHAOS
47
In [Zh], P. Zhao generalized the result in [LS, LS2] to Maass cusp forms. Let ϕj (z) be the
j-th Maass-Hecke eigenform, with the Laplacian eigenvalues 14 + rj2 .
Theorem 10.1. [Zh] Fix any ² > 0. Then we have
X −( rj −T )2
rj +T 2
(e T 1−² + e−( T 1−² ) )L(1, sym2 ϕj )µj (ϕ)µj (ψ)
tj
1
= T 1−² V (ϕ, ψ) + O(T 2 +² ),
V is diagonalized by the orthonormal basis {ϕj } of Maass-Hecke cusp forms and the eigenvalue of V at a Maass-Hecke cusp form ϕ is
1
L( , ϕ)VC (ϕ, ϕ).
2
It would be desirable to remove the weights. The unweighted version should be (P. Zhao,
personal communication; to appear)
X
(88)
µj (ϕ)µj (ψ) ∼ λV (ϕ, ψ).
λj ≤λ
10.3. General case. The only rigorous result to date which is valid on general Riemannian
manifolds with hyperbolic geodesic flow is the logarithmic decay:
Theorem 10.2. [Z4] For any (M, g) with hyperbolic geodesic flow,
1
1 X
|(Aϕj , ϕj ) − ω(A)|2p = O(
).
N (λ) λ ≤λ
(log λ)p
j
It is again based on Ratner’s central limit theorem for the geodesic flow. The logarithm
reflects the exponential blow up in time of remainder estimates for traces involving the wave
group associated to hyperbolic flows and thus the necessity of keeping the time less than the
Ehrenfest time (18). It would be surprising if the logarithmic decay is sharp for Laplacians.
However, R. Schubert shows in [Schu, Schu2] that the estimate is sharp in the case of twodimensional hyperbolic quantum cat maps. Hence the estimate cannot be improved by
semi-classical arguments that hold in both settings.
R
10.4. Variance and Patterson-Sullivan distributions. When S ∗ XΓ adµL = 0, the LuoSarnak variance results for Hecke eigenfunctions have the form,
X ¯
¯
1
¯ha, dWir i¯2 ∼ 1 V (a, a).
(89)
j
N (λ)
λ
j:|rj |≤λ
Even if one knew that asymptotics of this kind should occur, it does not seem apriori obvious
that the bilinear form on the right side should be invariant under g t because V (a, a) occurs
in the λ1 term in the variance asymptotics and only only knows that the Wigner distributions
are invariant modulo terms of order λ1j . A quick way to see that the bilinear form must be
g t -invariant is via the Patterson-Sullivan distributions. From (51), we have
X
X
1
1
(90)
|ha, dWirj i|2 =
|ha, dPˆS irj i|2 + O(λ−2 ).
N (λ)
N (λ)
j:|rj |≤λ
j:|rj |≤λ
48
STEVE ZELDITCH
P
Indeed, the difference is of the form N 1(λ) j:|rj |≤λ r2j Re ha, dPˆS irj i · hL2 a, dPˆS irj i + O(λ−2 ).
The sum in the last expression is also of order O(λ−2 ) (e.g. one can go back and express each
factor of P S with one of Wirj and apply the Luo-Sarnak asymptotics again). Since PˆS irj is
g t -invariant, it follows that V (a, a) must be g t invariant. This argument is equally valid on
any compact hyperbolic surface, but of course there is no proof of such asymptotics except in
the arithmetic Hecke case. It would be interesting to draw further relations between V (a, a)
and P Sirj .
11. Entropy of quantum limits on manifolds with Anosov geodesic flow
So far, the results on quantum limits have basically used the symmetry (or invariance)
properties of the limits. But generic chaotic systems have no symmetries. What is there to
constrain the huge number of possible limits?
The recent results of Anantharaman [A], Anantharaman- Nonnenmacher [AN] and Rivière
[Riv], give a very interesting answer to this question (see also [ANK]) for (M, g) with Anosov
geodesic flow. They use the quantum mechanics to prove lower bounds on entropies of
quantum limit measures. The lower bounds eliminate many of the possible limits, e.g. they
disqualify finite sums of periodic orbit measures.
The purpose of this section is to present the results of [A] and [AN] in some detail. Both
articles are based on a hyperbolic dispersive estimate (called the main estimate in [A])
which, roughly speaking, measures the norm of quantized cylinder set operators in terms
of ~ and the length of the cylinder. But they use the estimate in quite different ways. In
[A], it is used in combination with an analysis of certain special covers of S ∗ M by cylinder
sets that are adapted to the eigenfunctions. One of the important results is an estimate
on the topological entropy htop (supp(ν0 )) of the support of any quantum limit measure. In
[AN] the key tool is the “entropic uncertainty principle”. It leads to a lower bound for the
Kolomogorv-Sinai entropy hKS (ν0 ) of the quantum limit.
There now exist several excellent and authoritative expositions of the KS entropy bounds
and the entropic uncertainty principle [A2, ANK, AN2, CV3] in addition to the well written
initial articles [A, AN] (which take considerable care to give intuitive explanations of technical
steps). For this reason, we emphasize the approach in [A]. We closely follow the original
references in discussing heuristic reasons for the lower bounds and outlining rigorous proofs.
We also discuss earlier entropy lower bounds of Bourgain-Lindenstrauss [BL] and Wolpert
[W3] in the case of Hecke eigenfunctions.
Before stating the results, we review the various notions of entropy.
11.1. KS entropy. We first recall the definition of the KS entropy of an invariant probability measure µ for the geodesic flow. Roughly speaking, the entropy measures the average complexity of µ-typical orbits. In the Kolmogorov-Sinai entropy, one starts with a
partition P = (E1 , . . . , Ek ) of Sg∗ M and defines the Shannon entropy of the partition by
P
hP (µ) = kj=1 µ(Ej ) log µ(Ej ). Under iterates of the time one map g of the geodesic flow,
one refines the partition from the sets Ej to the cylinder sets of length n:
P n := {Pαn := Eα0 ∩ g −1 Eα1 ∩ · · · ∩ g −n+1 Eαn−1 : [α0 , . . . , αn−1 ] ∈ Nn }.
One defines hn (P, µ) to be the Shannon entropy of this partition and then defines hKS (µ, P) =
limn→∞ n1 hn (µ, P). Then hKS (µ) = supP hKS (µ, P).
RECENT DEVELOPMENTS IN MATHEMATICAL QUANTUM CHAOS
49
The measure µ(Eα0 ∩ g −1 Eα1 ∩ · · · ∩ g −n+1 Eαn−1 ) is the probability with respect to µ that
an orbit successively visits Eα0 , Eα1 , · · · . The entropy measures the exponential decay rate
of these probabilities for large times.
11.2. Symbolic coding and cylinder sets. In the (M, g) setting, we fix a partition {Mk }
of M and a corresponding partition T ∗ Mk of T ∗ M . Let Pαk be the characteristic function
of T ∗ Mk . (Later it must to be smoothed out). Let Σ = {1, . . . , `}Z where ` is the number of
elements of the partition P 0) . To each tangent vector v ∈ S ∗ M , one can associated a unique
element I(v) = (αk ) ∈ Σ so that g n v ∈ Pαj for all n ∈ Z. This gives a symbolic coding map
I : S ∗ M → Σ. The time one map g 1 then conjugates under the coding map to the shift
σ((αj )) = ((αj+1 )) on admissible sequences, i.e. sequences in the image of the coding map.
An invariant measure ν0 thus corresponds to a shift invariant measure µ0 on Σ.
A cylinder set [α0 , . . . , αn−1 ] ⊂ Σ of length n is the subset of Σ formed by sequences with
the given initial segment. The set of such cylinder sets of length n is denoted Σn . Cylinder
sets for Σ are not the same as cylinder sets for g 1 , which have the form Pα0 ∩ g −1 Pα1 ∩ · · · ∩
g −n Pαn .
The µ0 measure of a Σ-cylinder set is by definition,
µ0 ([α0 , . . . , αn ]) = ν0 (Pα0 ∩ g −1 Pα1 ∩ · · · ∩ g −n+1 Pαn−1 ).
11.3. Bowen balls. Cylinder sets are closely related to Bowen balls, i.e. balls in the Bowen
metric. For any smooth map f : X → X, the Bowen metric at time n is defined by
dfn (x, y) = max d(f i x, f i y).
0≤i≤n−1
Let Bnf (x, r) be the r ball around x in this metric.
In the continuous time case of the geodesic flow g t , one defines the dT metric
(91)
dT (ρ, ρ0 ) = max d(g t (ρ), g t (ρ0 )).
t∈[0T ]
The Bowen ball BT (ρ, r) at time T is the r- ball in this metric.
As with cylinder sets, if ρ, ρ0 lie in BT (ρ, ²) then their orbits are ²-close for the interval
[0, T ]. This is not quite the same as running through the same elements of a partition but
for large T the balls and cylinders are rather similar. This is because the geodesic flow g t
stretches everything by a factor of et in the unstable direction, and contracts everything
by e−t in the stable direction; it preserves distances along the geodesic flow. In variable
curvature and in higher dimensions, the ‘cube’ becomes a rectangular parallelopiped whose
axes are determined by the Lyapunov exponents.
In the case of geodesic flows of compact hyperbolic manifolds of constant curvature −1,
Bowen balls BT (ρ, r) are roughly of radius re−T in the unstable direction, and r in the stable
direction and geodesic flow directions. In the hyperbolic case of G/Γ with co-compact Γ, the
tube (in the notation of §1.15.1) is
(92)
BT (ρ, ²)) = a((−r, r))n− (r, r))n+ ((−e−T r, e−T r)).
To make the ball symmetric with respect to the stable and unstable directions is to make
it symmetric with respect to time reversal. One uses the time interval [−T /2, T /2] instead
of [0, T ] and defines the new distance,
(93)
d0T (ρ, ρ0 ) =
max
t∈−[T /2,T /2]
d0 (g t (ρ), g t (ρ0 )),
50
STEVE ZELDITCH
We then denote the r ball by BT0 (ρ, r). In the hyperbolic case,
(94)
BT0 (ρ, ²)) = a((−², ²))n− ((−e−T /2 ², e−T /2 ²))n+ ((−e−T /2 ², e−T /2 ²)).
Thus, the Bowen ball in the symmetric case (in constant curvature) is a ball (or cube) of
radius re−T in the transverse direction to the geodesic flow. The length along the flow is not
important.
11.4. Brin-Katok local entropy. Kolmogorov-Sinai entropy of an invariant measure is
related to the local dimension of the measure on Bowen balls. This is the approach in
Bourgain-Lindenstrauss [BL] and in [LIND], and stems from work of Young and LedrappierYoung.
Define the local entropy on an invariant measure µ for a map f by
− log µ(Bnf (x, δ))
− log µ(Bnf (x, δ))
hµ (f, x) = lim lim sup
= lim lim inf
.
δ→0 n→∞
δ→0 n→∞
n
n
Brin-Katok proved that both limits exist, that the local entropy hµ (f, x) is f invariant
and
Z
hµ (f ) =
hµ (f, x)dµ(x).
X
The definition and result is similar in the geodesic flow case. The local entropy of µ at ρ
is defined by,
lim lim inf −
²→0 T →∞
1
1
log µ(BT (ρ, ²)) = lim lim sup − log µ(BT (ρ, ²)) := hKS (µ, ρ).
²→0 T →∞
T
T
11.5. Ergodic decomposition. Let ν be an invariant measure. It may be expressed as a
convex combination of ergodic measures, which are extreme points of the compact convex
set of invariant measures. There is a concrete formula,
Z
(95)
ν = νxE dν(x)
for its ergodic decomposition, where νxE is the orbital average through x ∈ G/Γ. It is a fact
that νxE ergodic for ν-a.e. x. Then the local entropy is
µ̃(T (x, ²))
, a.s.µ̃.
²→0
log ²
ha(t) (µ̃Ex ) = lim
(96)
The Brin-Katok theorem states that the global KS entropy is
Z
h(ν) = h(νxE )dν(x).
11.6. Topological entropy of an invariant subset. Anantharaman’s first entropy bound
refers to the topological entropy of an invariant set F . Then, by definition,
htop (F ) ≤ λ
⇐⇒
∀δ > 0, ∃C > 0 : F can be covered by at most
(97)
Cen(λ+δ) cylinders of length n, ∀n.
RECENT DEVELOPMENTS IN MATHEMATICAL QUANTUM CHAOS
51
11.7. Bourgain-Lindenstrauss entropy bound. Before discussing the work of AnantharmanNonnenmacher, we review a simpler entropy bound of Bourgain-Lindenstrauss. After initial
work of Wolpert [W3], Bourgain-Lindenstrauss [BL] obtained a strong lower bound on entropies of quantum limit measures associated to Hecke eigenfunctions. Although it was later
surpassed by Lindenstrauss’ QUE theorem [LIND], it is simple to state and illustrates the
use of the local entropy and local dimensions to measure the KS entropy.
The setting is a compact or finite area arithmetic hyperbolic quotient. We have not
reviewed local entropy in the non-compact finite area case, but proceed by analogy with the
compact case. Then
B(², τ0 ) == a((−τ0 , τ0 ))n− (², ²))n+ ((−², ²))
is a Bowen ball around the identity element and xB(², τ0 ) is its left translate by x ∈ G. Here
² = re−T /2 .
Theorem 11.1. [BL] Let Γ ⊂ SL(2, Z) be a congruence subgroup. Let µ be a quantum limit.
Then for any compact subset K ⊂ Γ\G and any x ∈ K,
µ(xB(², τ0 )) ≤ ²2/9 .
In the flow-box notation above,
(98)
µn (BT (ρ, ²)) ≤ Ce−T /9 .
This implies that any ergodic component of any quantum limit has entropy ≥ 91 .
Corollary 11.2. [BL] Almost every component of a quantum limit measure has entropy
≥ 1/9. The Hausdorff dimenson of the support of the limit measure is ≥ 1 + 1/9.
Theorem 11.1 is a consequence of the following uniform upper bound for masses in small
tubes around geodesic segments in configuration space:
Theorem 11.3. [BL] Let Γ ⊂ SL(2, Z) be a congruence subgroup. Let µ be a quantum limit.
Then for any compact subset K ⊂ Γ\G and any x ∈ K, and for any Hecke eigenfunction
ϕλj ,
Z
xB(r,τ0 )
|ϕλj |2 dV ≤ r2/9 .
One may rewrite this result in terms of a Riesz-energy of the quantum limit measures:
Corollary 11.4. Let µ be a quantum limit of a sequence of Hecke-Maass eigenfunctions.
Then for κ < 2/9,
Z Z
dµ(x)dµ(y)
< ∞.
κ+1
M M dM (x, y)
Although these results were superseded by Lindenstrauss’ QUE bound, we briefly sketch
the idea of the proof since it makes an interesting contrast to the Anantharaman-Nonnenmacher
bound. They prove that there exists a set W of integers of size ²−2/9 so that for n ∈ W one
has (roughly) that the translates yB(², τ0 ) of the small balls by y in the Hecke correspondence
for Tn are all pairwise disjoint. Since
X
|ϕλ (y)|2 ≤
|ϕλ (z)|2 ,
z∈Tn (y)
52
one has
STEVE ZELDITCH
R
|ϕλ (y)|2 dV
xB(²,τ0 )
≤ C
=
R
P
xB(²,τ0 )
z∈Tn (y)
R
P
z∈Tn (x)
zB(²,τ0 )
|ϕλ (z)|2 dV
|ϕλ (z)|2 dV.
Now sum over n ∈ W and use the disjointness of the small balls zB(², τ0 ) and the L2
normalization of the eigenfunction to obtain
R
R
P
P
|ϕλ (y)|2 dV ≤ |W1 | n∈W z∈Tn (x) zB(²,τ0 ) |ϕλ (z)|2 dV
xB(²,τ0 )
≤ C²2/9
R
X
|ϕ2λ dV ≤ C²2/9 .
11.8. Anantharman and Anantharaman-Nonnenacher lower bound. We now turn
to the lower bounds on entropy in the general Anosov case given in [A] and in [AN, ANK].
The mechanisms proving the lower bounds and the results are somewhat different between
the two articles. The result of [A] gives a lower bound for the topological entropy htop (suppν0 )
of the support of a quantum limit while the main result of [AN] gives a lower bound on the
metric or Kolmogorov Sinai entropy hg = hKS of the quantum limits. For the definition of
htop (F ), see (97).
R
Theorem 11.5. [A] Let ν0 = S ∗ M ν0x dν0 (x) be the ergodic decomposition of ν0 . Then there
exists κ > 0 and two continuous decreasing functions τ : [0, 1] → [0, 1] and ϑ : (0, 1] → R+
with τ (0) = 1, ϑ(0) = ∞, such that
¾
½
κ 2
Λ
x
) (1 − τ (δ)).
ν0 [ x : hg (ν0 ) ≥ (1 − δ) ] ≥ (
2
ϑ(δ)
Hence
• hg (ν0 ) > 0;
• htop (supp ν0 ) ≥ Λ2 .
It follows that a positive proportion of the ergodic components of ν0 must have KS entropy
arbitrarily close to Λ2 .
This is proved as a consequence of a Proposition which may have other applications:
Theorem 11.6. [A] (Proposition 2.0.4) With the same notation as in Theorem 11.5 and
Remark 11.10. Let F be a subset with htop (F ) ≤ Λ2 (1 − δ). Then,
µ
¶
κ 2
ν0 (F ) ≤ 1 − (
) (1 − τ (δ)) < 1.
ϑ(δ)
Hence supp ν0 cannot equal F .
In [A] Theorem 1.1.2, there is a generalization to quasi-modes which introduces a constant
c. For simplicity we only consider eigenfunctions.
In the subsequent article of Anantharaman-Nonnenmacher, [AN, ANK] the authors obtain
a quantitative lower bound on the KS entropy:
Theorem 11.7. [AN, ANK] Let µ be a semiclassical measure associated to the eigenfunctions of the Laplacian on M . Then its metric entropy satisfies
¯
¯Z
¯ (d − 1)
¯
u
¯
log J (ρ)dµ(ρ)¯¯ −
(99)
hKS (µ) ≥ ¯
λmax ,
2
∗
S M
RECENT DEVELOPMENTS IN MATHEMATICAL QUANTUM CHAOS
53
where d = dim M . λmax = lim|t|→∞ 1t log supρ∈E |dgρt | is the maximal expansion rate of the
geodesic flow on E (5) and J u is the unstable Jacobian determinant (3).
When M has constant sectional curvature −1, the theorem implies that
d−1
.
(100)
hKS (µ) ≥
2
They state the conjecture
Conjecture 11.8. [AN] Let (M, g) have Anosov geodesic flow. Then for any quantum
limit measure ν0 ,
¯Z
¯
¯
1 ¯¯
u
hg (ν0 ) ≥ ¯
log J (v)dν0 (v)¯¯ .
2 S∗M
This conjecture has recently been proved by G. Rivière for surfaces of negative curvature
[Riv].
This conjecture does not imply QUE. In fact the counterexamples in [FNB, FN] to QUE
in the case of “quantum cat maps” satisfy the condition of the conjecture. So there is some
chance that the conjecture is best possible for (M, g) with ergodic geodesic flow.
To the author’s knowledge, the corollary that quantum limits in the Anosov case cannot
be pure periodic orbit measures has not been proved a simpler way that by applying the
theorems above. In §11.21 we will go over the proof in that special case and see how it ties
together with mass concentration of eigenfunctions around hyperbolic closed geodesics in §5.
11.9. Problem with semi-classical estimates and Bowen balls. As a first attempt to
estimate KS entropies of quantum limits, one might try to study the local entropy formula
as in Bourgain-Lindenstrauss. Let BT (ρ, ²) be the small tube around ρ ∈ S ∗ M , i.e. of length
² in the flow direction and e−T /2 in the stable/unstable directions. As in the BourgainLindenstrauss bound, one would like to understand the decay of µn (BT (ρ, ²)) as T → ∞ and
n → ∞. In the brief expository article [AN2], the authors give the heuristic estimate
(101)
d−1
µn (BT (ρ, ²)) ≤ Cλn 2 e−
(d−1)T
2
, d = dim M
on the local dimensions of the quantum measures for a sequence of eigenfunctions. Unlike
the uniform Bourgain-Lindenstrauss estimate, the estimate is λn -dependent and is only nontrivial for times T > log λn . But then the ball radius (in the stable/unstable directions) is
−1/2
e−T = λn , which is just below the minimal scale allowed by the uncertainty principle (§1.4).
To obtain a useful entropy bound one needs to take T = 2 log λn and then e−T = λ−1
n , well
below the Heisenberg uncertainty threshold. So although it is attractive, the local entropy
is difficult to use. Recall that Bourgain-Lindenstrauss used the many elements of the Hecke
correspondences to move the small ball around so that it almost fills out the manifold, and
then used the L2 normalization to estimate the total mass. There doesn’t seem to exist a
similar mechanism to build up a big mass from the small balls in the general Anosov case.
11.10. Some important parameters. Two important parameters κ, ϑ appear here and
in the statements and proofs in [A]. They represent time scales relative to the Ehrenfest
time:
• κ: Semi-classical estimates only work when n ≤ κ| log ~|, i.e. κ| log ~| plays the role
of the Ehrenfest time.
54
STEVE ZELDITCH
• ϑ: The main estimate is only useful when h−d/2 e−nΛ/2 << 1 or n ≥ ϑ| log ~|.
11.11. Cylinder set operators in quantum mechanics. To get out of the too-small
Bowen ball impasse, Anantharaman [A] and Ananthraman-Nonnenmacher study “quantum
measures” of cylinder sets C rather than small balls. The emphasis (and results) on quantum cylinder sets is one of the key innovations in [A]. The classical cylinder sets are not
directly involved in the main estimates of [A, AN]. Rather they are quantized as cylinder set
operators. Then their “quantum measure” or quantum entropy is studied. To distinguish
notationally between classical and quantum objects, we put a ˆ· on the quantum operator.
To define quantum cylinder set operators, one quantizes the partition to define a smooth
quantum partition of unity P̂k by smoothing out the characteristic functions of Mk . Let
α = [α0 , . . . , αn−1 ∈ Σn be a cylinder set in sequence space Σ. The corresponding quantum
cylinder operator is
(102)
P̂α = P̂αn−1 (n − 1)P̂αn −2 (n − 2) · · · P̂α0 ,
where α = [α0 , . . . , αn−1 ] where
P̂ (k) = U ∗k P̂ U k .
(103)
√
Here, U = ei ∆ is the propagator at unit time (or in the semi-classical framework, U = ei~∆/2
(see [A] (1.3.4)).
The analogue of the measure of a cylinder set in quantum mechanics is given by the
matrix element of the cylinder set operator in the energy state, transported to Σ. We use
the semi-classical notation of [A], but it is easily converted to homogeneous notation.
Definition:
(See [A], (1.3.4)) Let ψ~ be an eigenfunction of ∆. Define the associated
quantum “measure” of cylinder sets C = [α0 , . . . , αn−1 ] ∈ Σn by
(104)
µ~ ([α0 , . . . , αn ]) = hP̂αn (n) · · · P̂α0 (0)ψ~ , ψ~ i.
Thus, one “quantizes” the cylinder set C = [α0 , . . . , αn−1 ] as the operator
Cˆ = U −(n−1) P̂αn−1 U P̂αn−2 · · · U P̂α0
(105)
For each eigenfunction, one obtains a linear functional
ˆ
µ~ ([α0 , . . . , αn−1 ]) = ρ~ (C)
(106)
in the notation of states of §3. The ‘quantum measures’ are shift invariant, i.e.
µ~ ([α0 , . . . , αn−1 ]) = µ~ (σ −1 [α0 , . . . , αn−1 ]).
This says,
hU −(n−2) Pαn−1 U n−2 · · · U Pα0 U −1 ϕ~ , ϕ~ i = hU −(n−1) Pαn−1 U n−1 · · · U Pα0 ϕ~ , ϕ~ i,
which is true since ρh (A) = hAϕ~ , ϕ~ i is an invariant state.
The matrix element hP̂α ψ~ , ψ~ i is the probability (amplitude) that the particle in state ψ~
visits the phase space regions Pα0 , Pα1 , . . . , Pαn−1 at times 0, 1, . . . , n − 1.
Thus, the states ρ~ or alternatively the Wigner distributions of ψ~ are transported to Σ
to define linear functionals µ~ on the span of the cylinder functions on Σ. They exactly
invariant under the shift map.
RECENT DEVELOPMENTS IN MATHEMATICAL QUANTUM CHAOS
55
The functionals µ~ are not positive measures since Cˆ~ is not a positive operator. However
as in the proof of the quantum ergodicity Theorem 4.1, one has the Schwartz inequality,
ˆ 2 ≤ ρ~ (Cˆ∗ C),
ˆ
(107)
|ρ~ (C)|
reflecting the positivity of the state ρ~ .
11.12. Main estimate for cylinder sets of Σ. The so-called main estimate of [A] implies
that for every cylinder set C ∈ Σn , one has
(108)
|µ~ (C)| ≤ Cβ h−d/2 e−nΛ/2 (1 + O(²))n ,
uniformly for n ≤ β | log ~|.
Here, β can be taken arbitrarily large. This estimate is similar in spirit to (101) but circumvents the uncertainty principle. It is one of the essential quantum mechanical (or semiclassical) ingredient for getting a lower bound on htop (suppν0 ) of supports of quantum limit
measures (see §11.15).
11.13. Quantization of cylinder sets versus quantized cylinder operators. Let us
explain how (108) gets around the uncertainty principle. A fundamental feature of quantization is that it is not a homomorphism. Thus, the quantization of the (smoothed) characteristic function of the cylinder set Pα0 ∩ g −1 Pα1 ∩ · · · ∩ g −n Pαn is very different from
P̂αn−1 U P̂αn−2 · · · U P̂α0 , the ordering which occurs in the quantum cylinder operator P̂α in
(102). The problematic Bowen ball estimate (101) would arise if one quantized the Bowen
ball (or cylinder set) in the first sense and then took its matrix element. Quantizing in the
other order makes the main estimate possible.
11.14. Quantized cylinder operators and covering numbers. As motivation to study
quantum measures of cylinder sets, Anantharaman observes that the main estimate suggests
an obstruction to a quantum measure concentrating on a set F with htop (F ) < Λ2 . For each n
Λ
such a set can be covered by Cen( 2 (1−δ) cylinder sets of length n. By the main estimate (108),
each of the cylinder sets has quantum measure |µ~ (C)| ≤ Ce−d/2 log ~−nΛ/2 uniformly for n ≤
κ | log ~| for any κ. As yet, the combination does not disprove that suppν0 = F, i.e. ν0 (F ) =
1, since the the large factor of h−d/2 is not cancelled in the resulting estimate µ~ (F ) ≤
Λ
Cen( 2 (1−δ)−d/2 log ~−nΛ/2 . An additional idea is needed to decouple the large time parameter
n into independent time parameters. Roughly speaking, this is done by a sub-multiplicative
property. In §11.20, the ideas in the above paragraph are made more precise and effective.
11.15. Upper semi-continuity, sub-multiplicativity and sub-additivity. Before getting into the details of the (sketch of) proof, it is useful to consider why there should exist
a non-trivial lower bound on hKS (ν0 ) for a quantum limit measure? What constrains a
quantum limit measure to have higher entropy than some given invariant measure?
First, the classical entropy of an invariant is upper semi-continuous with respect to weak
convergence. This is in the right direction for proving lower entropy bounds for quantum
limits. If one can obtain a lower bound on the ‘entropies’ of the quantum measures µn ,
then the entropy of the limit can only jump up. Of course, the quantum measures are not
g t invariant measures and so upper semi-continuity does not literally apply. One needs to
define a useful notion of entropy for the quantum measures and prove some version of upper
semi-continuity for it. The USC property of the classical entropy suggests that there should
exist some analogue on the quantum level.
56
STEVE ZELDITCH
In [A], entropies are studied via special covers by cylinder sets. For these, there is a
sub-multiplicative estimate on the number of elements in the cover.
In [AN], a quite different idea is used: the authors employ a notion of quantum entropy
due to Maassen-Uffink and prove a certain lower bound for the quantum entropy called the
entropy uncertainty inequality. It is the origin of the lower bound for the quantum entropies
and the KS entropy in Theorem 11.7 (see §11.17). As a replacement for the USC property
of the entropy, a certain sub-additivity property for the quantum entropy is proved (see
§11.18).
In the next three subsections, we go over the main ingredients in the proof. Then we give
a quick sketch of the full proof.
11.16. Main estimate in more detail. The main estimate in [A] (Theorem 1.3.3) was
refined in [AN, ANK]. To state the results, we need some further notation. Let nE (~) denote
the Ehrenfest time
·
¸
(1 − δ)| log ~|
nE (~) =
λmax
where the brackets denote the integer part and δ is a certain small number arising in energy
localization (for the definition of λmax see (5)). The authors also introduce a discrete ‘coarsegrained’ unstable Jacobian
J1u (α0 , α1 ) := sup{J u (ρ) : ρ ∈ T ∗ Ωα0 ∩ E ² : g t ρ ∈ T ∗ Ωα1 },
for α0 , α1 = 1, . . . , K. Here, Ωj are small open neighborhoods of the partition sets Mj . For
a sequence α = (α0 , . . . , αn−1 ) of symbols of length n, one defines
Jnu (α) := J1u (α0 , α1 ) · · · J1u (αn−2 , αn−1 ).
Theorem 11.9. (See Theorem 3.5 of [ANK]) Given a partition P (0) and δ, δ 0 > 0 small
enough, there exists ~P (0) ,δ,δ0 such that, for any ~ ≤ ~P (0) ,δ,δ0 , for any positive integer n ≤
nE (~), and any pair of sequences α, α0 of length n,
¯
¯
p
¯ ∗ n
(n) ¯
(109)
¯|P̂α0 U P̂α Op(χ )¯ | ≤ C ~−(d−1+cδ) Jnu (α)Jnu (α0 ) .
Here, d = dim M and the constants c, C only depend on (M, g).
There is a more refined version using a sharper energy cutoff in [AN].
Theorem 11.10. Given a partition P (0) , κ > 0 and δ > 0 small enough, there exists
~P (0) ,δ,κ such that, for any ~ ≤ ~P (0) ,δ,κ , for any positive integer n ≤ κ| log ~|, and any pair
of sequences α, α0 of length n,
¯
¯
d−1
¯ ∗ n
(n) ¯
(110)
¯|P̂α0 U P̂α Op(χ )¯ | ≤ C ~−( 2 +δ) e−(d−1)n (1 + O(~δ ))n .
Here, d = dim M .
To prove this, one shows that any state of the form Op(χ(∗) )Ψ can be decomposed as a
(d−1)
superposition of essentially ~− 2 normalized Lagrangian states, supported on Lagrangian
manifolds transverse to the stable leaves of the flow. The action of the operator P̂α on such
Lagrangian states is intuitively as follows: each application of U stretches the Lagrangian
in the unstable direction (the rate of elongation being described by the unstable Jacobian)
whereas each multiplication by P̂αj cuts off the small piece of the Lagrangian in the αth cell.
RECENT DEVELOPMENTS IN MATHEMATICAL QUANTUM CHAOS
57
This iteration of stretching and cutting accounts for the exponential decay. A somewhat
more detailed exposition is in §11.22.
This estimate can be reformulated as a statement about matrix elements of the operators
P̂α relative to eigenfunctions (see Theorem 3.1 of [ANK])
Theorem 11.11. Let P̂k = 1sm
Mk . For any κ > 0 there exists ~κ > 0 so that, uniformly for
~ < ~κ and all n ≤ κ | log ~|, and for all α = (α0 , . . . , αn−1 ) ∈ [1, κ]n ∩ Zn ,
Λ
||P̂α ψh || ≤ 2(2π~)−d/2 e−n 2 (1 + O(²))n .
The cutoff operator Op(χ(n) ) is supported near the energy surface λ2n , and the eigenfunctions are supported on their energy surfaces. Here, Λ is the “smallest” expansion rate (see
(4)).
We recall that hP̂α ψ~ , ψ~ i is the probability (amplitude) that the particle in state ψ~ visits
the phase space regions Pα0 , Pα1 , . . . , Pαn−1 at times 0, 1, . . . , n − 1. The main estimates show
that this probability decays exponentially fast with n, at rate Λ2 . However, the decay only
~|
starts near the Eherenfest time n1 = d| log
.
Λ
11.17.
Entropy uncertainty principle. The source of the lower bound for hKS (ν0 )
is most simply explained in [AN]. It is based on the ‘entropic uncertainty principle’ of
Maassen- Uffink. There are several notions of quantum or non-commutative entropy, but for
applications to eigenfunctions it is important to find one withp
good semi-classical properties.
Let (H, h., .i) be a complex Hilbert space, and let ||ψ|| = hψ, ψi denote the associated
norm. The quantum notion of partition is a family π = (πk )k=1,...,N of operators on H such
P
∗
that H
k=1 πk πk = Id. If ||ψ|| = 1, the entropy of ψ with respect to the partition π is define
by
N
X
||πk∗ ψ||2 log ||πk∗ ψ||2 .
hπ (ψ) = −
k=1
We note that the quantum analogue of an invariant probability measure
P µ is an invariant
state ρ, and the direct analogue of the entropy of the partition would be ρ(πk πk∗ ) log ρ(πk πk∗ ).
If the state is ρ(A) = hAψ, ψi then ρ(πk πk∗ ) = ||πk∗ ψ||2 .
The dynamics is generated by a unitary operator U on H. We now state a simple version
of the entropy uncertainty inequality of Maasen-Uffink. A more elaborate version in [A, AN,
ANK] gives a lower bound for a certain ‘pressure’.
Theorem 11.12. For any ² ≥ 0, for any normalized ψ ∈ H,
¡ ¢
¡ ¢
hπ Uψ + hπ ψ ≥ −2 log c(U) ,
where
c(U) = sup |hek , Uej i|
j,k
is the supremum of all matrix elements in the orthonormal basis {ej }. In particular, hπ (ψ) ≥
− log c(U) if ψ is an eigenfunction of U.
The intuitive idea is that if a unitary matrix has small entries, then each of its eigenvectors
must have large Shannon entropy.
58
STEVE ZELDITCH
This uncertainty principle is applied
to the entropies of the partitions defined above. For
√
λ
iTE ∆
the propagator, we put U = e
is the wave operator at the ‘Ehrenfest time’ TE = λlog
.
max
1
2
Or in the semi-classical framework (where h = λ ), one can use the Hamiltonian is H = h ∆
log
1
and the time evolution U = einE (h)h∆ with nE (h) = λmaxh .
Applied to an eigenfunction of U, one has
X
(111)
hπ (ψh ) =
||P̂α∗ ψh ||2 log ||P̂α∗ ψh ||2 ,
|α|=nE
and obviously
¡ ¢
¡ ¢
¡ ¢
hπ Uψ + hπ ψ = 2hπ ψ .
On the right side,
(112)
c(U) =
max
0
|α|=|α |=nE
||P̂α0 U nE P̂α Op(χ(nE ) )||
where χ(nE ) is a very sharp energy cutoff supported in a tubular neighborhood E ² := H −1 (1−
², 1 + ²) of E = S ∗ M of width 2h1−δ enδ for a given δ > 0.
11.18. Sub-additivity. Another key component in the proof is that the quantum entropy
is almost sub-additive for ~ ≤ | log t|.
Sub-additivity of the classical KS entropy follows from the concavity of the function
−x log x. It states that the sequence {hn (µ, P )} is sub-additive, i.e.
hn+m (µ, P ) ≤ hn (µ, P ) + hm (µ, P ).
However, this is false for the quantum measures.
The correct statement is as follows: There exists a function R(n0 , ~) such that lim~→0 R(n0 , ~)
log ~|
for all n0 ∈ Z and such that for all n0 , n ∈ N with n0 +n = (1−δ)|
, and for any normalized
λmax
eigenfunction ϕλ , one has
(113)
hP n0 +n (ϕλ ) ≤ hP n0 (ϕλ ) + hP n (ϕλ ) + R(n0 , ~).
11.19. Outline of proof of Theorem 11.7. We outline the proof from [AN]. Let
X
||P̂α∗ ψ||2 log ||P̂α∗ ψ||2 ,
(114)
hP (n) (ψ) = −
|α|=n
and let
(115)
hT (n) (ψ) = −
X
||P̂α ψ||2 log ||P̂α ψ||2 ,
|α|=n
(1) Let K =
1−δ
.
λmax
By the main estimate and the entropy uncertainty inequality, one has
hT (n) (ψh ) + hP (n) (ψh ) ≥ 2(d − 1)n +
(d − 1 + 2δ)λmax
nE (~) + O(1).
(1 − δ)
(2) Fix n0 , nE (~). Write n = qn0 + r, r ≤ n0 . Then by sub-additivity,
h (n ) (ψh ) hP (r) (ψh ) R(n0 , ~)
hP (n) (ψh )
≤ P 0
+
+
.
n
n0
nE (~)
n0
RECENT DEVELOPMENTS IN MATHEMATICAL QUANTUM CHAOS
59
(3) As ~ → 0, this gives
hP (n0 ) (ψh ) hT (n0 ) (ψh )
R(n0 , ~)
(d − 1 + 2δ)λmax
+
≥ 2(d − 1)n +
n + O(1) −
+ On0 (1/n).
n0
n0
(1 − δ)
2n0
(4) Now take the limit ~ → 0. The expressions ||P̂α ϕλ ||2 and the Shannon entropies tend
to their classical values. Hence, hP (n0 ) (ψh ) tends to the KS entropy of the quantum
limit. Thus, the lower bound implies
(d − 1 + 2δ)λmax
hn0 (µ)
≥d−1−
.
n0
2(1 − δ)
Since δ, δ 0 are arbitrary one can set them equal to zero. Then let n0 → ∞ to obtain
the KS entropy of the quantum limit.
11.20. Sketch of proof of Theorem 11.5 [A]. The proof of Theorem 11.5 is less structural than the proof of Theorem 11.7. There is no magic weapon like the entropic uncertainty
principle. In its place there is the construction of special covers by cylinder sets and counting
arguments for the number of elements in the cover. These covering arguments probably have
further applications and (although quantum) seem more geometric than the argument based
on the entropic uncertainty principle.
We mainly discuss the statement that htop (supp ν0 ) ≥ Λ2 . We recall (§11.6) that an inΛ
variant set F is with htop (F ) ≤ Λ2 (1 − δ) can be covered by at CeN ( 2 (1−δ) cylinder sets of
length N . We further recall that, by the main estimate (108), each of the cylinder sets has
quantum measure |µ~ (C)| ≤ Ce−d/2 log ~−N Λ/2 uniformly for N ≤ ϑ | log ~| for any ϑ. We
want to combine these estimates to show that ν0 (F ) < 1 for any quantum limit ν0 . But we
cannot simply multiply the measure estimate for C in the cover times the number of cylinder
sets in the cover. The main estimate being fixed, the only things left to work with are the
covers.
As will be seen below, one needs to introduce special covers adapted to the eigenfunctions.
The covering number of such covers has a sub-multiplicative property. We motivate it by
going through a heuristic proof from [A]. We also add a new step that helps explain why
special covers are needed. See also §11.21 where the pitfalls are explored when F = γ (a
closed hyperbolic orbit).
11.20.1. Covers and times, I. Let F ⊂ Σ be a shift-invariant subset with htop (F ) ≤ Λ2 (1 − δ).
Let Wn ⊂ Σn be a cover of minimal cardinality of F by n cylinder-sets. Now introduce a
new time N >> n and define

 {N − cylinders [α0 , . . . , αN −1 ] such that
(116)
ΣN (Wn , τ ) :=
 #{j∈[0,N −n]:[αj ,...,αj+n−1 ]∈Wn }
≥ τ.
N −n+1
They correspond to orbits that spend a proportion ≥ τ of their time in Wn . In terms of the
(ergodic) Liouville measure (transported to Σ), for almost every point of S ∗ M (or its image
under I), the proportion of the total time its orbit spends in Wn is the relative measure of
(the union of the cylinder sets in) Wn . If this measure is smaller than τ , then ΣN (Wn , τ )
is a rare event as N → ∞. Its cylinders satisfy σ j (C) ∩ Wn = C for a proportion τ of
0 ≤ j ≤ N . If one imagines partitioning the elements of Wn into N cylinders, then only a
60
STEVE ZELDITCH
(relatively) sparse subset will go into ΣN (Wn , τ ). The next Lemma gives an upper bound
on that number:
Lemma 11.13. (see Lemma 2.3.1 of [A]) For all n0 there exist n ≥ n0 and N0 so that for
N ≥ N0 and τ ∈ [0, 1], the cover Wn of F of minimal cardinality satisfies,
#ΣN (Wn , τ ) ≤ eN
3Λδ
8
eN htop (F ) e(1−τ )N (1+n) log ` .
where ` is the number of elements in the partition.
If τ is chosen sufficiently close to 1, this implies:
Corollary 11.14. (See (2.3.1) of [A]) If F is an invariant set with htop (F ) <
then
Λ
#ΣN (Wn , τ ) ≤ eN ( 2 (1−δ)) e(1−τ )N (1+n) log ` .
Λ
(1
2
− δ),
We now re-write this estimate in terms of the semi-classical parameter ~. If τ is sufficiently
close to 1 and ε > 0 is sufficiently small, there exists ϑ > 0 so that for N ≥ ϑ| log ~|,
(117)
Λ
eN ( 2 (1−δ)) e(1−τ )N (1+n) log ` < (1 − ε)hd/2 eN Λ/2 .
Indeed, let α = −δ Λ2 + (1 − τ )(1 + n) log `. For τ very close to 1 it is negative. We then
choose ϑ so that h−αϑ < (1 − ε)hd/2 , i.e. so that |αϑ| > d/2.
By the main estimate (108) and (117), we obtain the following
Corollary 11.15. (see [A], (2.0.1))
For τ sufficiently close to 1 and ε close to zero, there exists ϑ > 0 so that for N ≥ ϑ| log ~|,
(118)
|µ~ (ΣN (Wn , τ ))| ≤ 1 − ε.
S
Here, µh (ΣN (Wn , τ )) means µh ( C∈ΣN (Wn ,τ )) C).
11.20.2. Heuristic Proof. We now present a heuristic proof of Theorem 11.5 from Section 2
of [A] that explains the need for some rather technical machinery introduced below.
We recall that µ~ is a shift (σ)-invariant measure on Σ. Let us temporarily pretend that
µ~ is also a positive measure. Then we would have for N ≥ ϑ| log ~|,
|µ~ (Wn )| = | N 1−n
PN −n−1
k=0
µ~ (σ −k Wn )|
¯ ³
´¯
P −n−1
¯
¯
= ¯µ~ N 1−n N
1
−k
σ Wn ¯
k=0
(119)
≤
P
C∈ΣN (Wn ,τ )
µ~ (C) + τ
P
C ∈Σ
/ N (Wn ,τ )
µ~ (C)
= µ~ (ΣN (Wn , τ )) + τ µ~ (ΣN (Wn , τ )c )
≤ (1 − τ ) (1 − ε) + τ.
Here we used (118), and that the sum of the µh measures of ΣN (Wn , τ ) and its complement
add to one. Since Wn is fixed, the weak convergence µ~ → µ0 implies,
(120)
|µ0 (Wn )| ≤ (1 − τ ) (1 − ε)) + τ < 1.
Since F ⊂ Wn , the same estimate applies to F . So F cannot be the support of µ0 . “QED”
RECENT DEVELOPMENTS IN MATHEMATICAL QUANTUM CHAOS
61
In the above steps, we use that σ −k Wn ⊂ ΣN (Wn , τ ) for k ≤ N − n − 1 and that
(121)
NX
−n−1
1
1 −k ≤ 1,
N − n k=0 σ Wn
NX
−n−1
1
1 −k ≤ τ on ΣN (Wn , τ )c .
N − n k=0 σ Wn
Remark: There are two very innovative ideas in this argument. It is somewhat reminiscent
of the averaging argument of Theorem 4.1, but there is no averaging over eigenvalues and no
appeal to ergodicity (since there is no reason why a weak* limit of µ~ should be an ergodic
measure). The first idea is to study the special observables P̂α . They are not expressed in the
form Op~ (a), i.e. in terms of a symbol, and it would be counter-productive to do so. Thus,
the argument of (130) takes place entirely on the quantum level and does not make use of
any properties of the weak* limit (which are unknown); the semi-classical limit is taken after
the inequality is proved. The key innovation is the third step, where the sets ΣN (Wn , τ ) are
introduced and the sum is broken up into one over ΣN (Wn , τ ) and one over its complement.
As is evident from (121), ΣN (Wn , τ )c arises naturally as an approximation to the τ -sublevel
P −n−1
set of N 1−n N
1σ−k Wn . But note again that the the decomposition occurs in ΣN and is
k=0
used the on the quantum level. This is very different from using symbols of the Ĉ and their
time averages to to define the corresponding subsets in S ∗ M !
The flaw in the argument is in the third line, where we pretended that µ~ was a positive
measure. We recall that µ~ (C) converges to µ0 (C), which is a positive measure. But this is
for a fixed C. But it is also close to being a probability measure for cylinder sets of length
≤ κ| log ~|. But (118) is only valid for N ≥ ϑ| log ~|. Since ϑ > κ, one cannot use them both
simultaneously (see §11.10 for κ, ϑ)
One can attempt to get around this problem using the positivity of the underlying state
ρ~ . Define the function
WnN
NX
−n−1
1
:=
1 −k : Σ → R+ .
N − n k=0 σ Wn
Then in the third line we rigorously have,
³
´1/2
(122)
|µ~ (WnN )| ≤ ρ~ (ŴnN )∗ (ŴnN )
= ||ŴnN Op~ ψ~ ||.
Here, ŴnN is the quantization as a cylinder set operator. We further denote by ΣN\
(Wn , τ )
ˆ
the operator equal to the sum of C for C ∈ ΣN (Wn , τ ). Then
ΣN\
(Wn , τ ) + ΣN\
(Wn , τ )c = I.
So (with Op~ (χ) the energy cutoff),
(123)
By (121),
||ŴnN Op~ (χ)ψ~ || ≤ ||ŴnN ΣN\
(Wn , τ )Op~ (χ)ψ~ || + ||ŴnN ΣN\
(Wn , τ )c Op~ (χ)ψ~ ||.


(Wn , τ )Op~ (χ)ψ~ || ≤ ||ΣN\
(Wn , τ )Op~ (χ)ψ~ ||
 ||ŴnN ΣN\

 ||Ŵ N Σ \
c
\ c
N (Wn , τ ) Op~ (χ)ψ~ || ≤ τ ||ΣN (Wn , τ ) Op~ (χ)ψ~ ||.
n
.
62
STEVE ZELDITCH
To complete the proof of the final equality, need an estimate of the form,
||ΣN\
(Wn , τ )Op~ (χ)ψ~ || ≤ 1 − θ.
(124)
We also need
||ΣN\
(Wn , τ )c Op~ ψ~ || ≤ 1 − ||ΣN\
(Wn , τ )Op~ ψ~ || + O(~).
(125)
The latter holds formally since ΣN\
(Wn , τ ), ΣN\
(Wn , τ )c are semi-classically complementary
projections. But the time N ≥ ϑ| log ~| may be too large for this semi-classical estimate.
The heuristic arguments are sufficiently convincing to motivate a difficult technical detour
in which a certain sub-multiplicativity theorem is used to reduce the N in the argument
from N ≥ ϑ| log ~| (above the Ehrenfest time) to N ≤ κ| log ~| (below the Ehrenfest time).
It is the analogue in [A] of the sub-additivity step in [AN], and its purpose is to allow the
argument above to use only cylinder sets of length κ| log ~| (or balls of radius >> ~1/2 )
where semi-classical analysis is possible. But one has to make sure that the covering number
estimates in Lemma 11.13 and (117) do not break down. One has to choose special covers
so that there are not too many elements in the cover of F . This step is another crucial
innovation in [A], and in some ways the subtlest one.
11.20.3. (~, 1 − θ, n)-covers. The sub-multiplicativity step requires a new concept and a new
parameter θ. Fix (~, n, θ) and consider a subset W ⊂ Σn .
Definition: Say that W is an (~, 1 − θ, n) − cover of Σ if
X
(126)
||
Cˆ~ Op~ (χ)ψ~ || ≤ θ.
C∈W c
c
Here, W is the complement of W ; W is an (~, 1 − θ, n) cover of Σ if in this quantum sense
the measure of W c is < 1 (it is called “small” in [A] but in the end small means < 1; See the
Remark at the end of §11.20.4 ). In other words, the union of the cylinder sets of W might
not cover all of Σ (think S ∗ M ), but the mass of the eigenfunctions ψ~ outside of the covered
part is rather small. In this quantum sense, it is intuitively a union of cylinder sets whose
measure is ≥ 1 − θ. By definition, if W is an (~, 1 − θ, n)− cover, then it is an (~, 1 − θ0 , n)−
cover for any θ0 > θ.
The covering number of such a cover is defined by
(127)
N~ (n, θ) = #{W : W is a (~, 1 − θ, n) − cover of Σ}.
The sub-multiplicative property is
Lemma 11.16. (sub-multiplicativity)
(128)
N~ (kn, kθ(1 + O(n~α ))) ≤ N~ (n, θ)k .
We will not attempt to explain this inequality but rather will point out its consequences.
One is a lower bound on growth of counting numbers of minimal (~, 1 − θ, N )-covers as N
grows (Lemma 2.2.6 of [A]):
Lemma 11.17. Given δ > 0, there exists ϑ so that:
Λ
δ
• (i) For N = ϑ| log ~|, N~ (N, θ) ≥ (1 − θ)eN 2 (1− 16 ) ;
κ
Λ
δ
• (ii) For N = κ| log ~|, κ ≤ ϑ, N~ (N, ϑκ θ) ≥ (1 − θ) ϑ eN 2 (1− 16 ) ;
RECENT DEVELOPMENTS IN MATHEMATICAL QUANTUM CHAOS
63
Estimate (ii) follows from (i) and sub-multiplicativity. The time in (ii) is now below the
Ehrenfest time. The sub-multiplicative property then has the following consequence:
Lemma 11.18. (See [A], (2.3.2); compare (126)) Let κ, ϑ have the meanings in §11.10. Let
Wn ⊂ Σn be a cover of F by n-cylinders of minimal cardinality; and let N = κ| log ~|. Then
for any θ < 1,
X
κ
||
Cˆ~ Op~ (χ)ψ~ || ≥ θ.
ϑ
c
C∈ΣN (Wn ,τ )
Proof. (Sketch) The inequality goes in the opposite direction from that in the definition of
(~, 1 −θ, N )-covers in (11.19). Hence the Lemma is equivalent to saying that ΣN (W, τ ) is not
a (~, 1 − ϑκ θ, N )-cover. This follows by comparing the upper bound in Lemma 11.13 with the
lower bound in 11.17. If ΣN (Wn , τ ) were such a cover, one would have ϑκ | log(1 − θ)| ≥ N N2Λ ,
which can only occur for finitely many N if θ < 1. The cardinality of ΣN (W, τ ) is too small
to be (~, 1 − θ, N )-cover when N = κ| log ~|.
¤
The lower bound is a key step in the proof of Theorem 11.5. Intuitively, it gives a lower
bound on eigenfunction mass outside the set of orbits which spend a proportion τ of their
time in Wn . The lower bound on the mass outside ΣN (Wn , τ ) eventually prohibits the
concentration of the quantum limit measure on F .
Using that Ĉ~ Op~ (χ) are roughly orthogonal projectors with orthogonal images for distinct
cylinders C, this leads to:
Lemma 11.19. (see [A], (2.4.5)) Continue with the notation of Lemma 11.18. For N =
κ| log ~|, we have
 P
δ

C∈ΣN |µ~ (C)| = 1 + O(~ ),




 P
¡ κ ¢2
|µ
(C)|
≥
θ + O(~κ ),
c
~
C∈ΣN (Wn ,τ )
ϑ




¡ ¢2

 =⇒ P
|µ (C)| ≤ 1 − κ θ + O(~κ ).
C∈ΣN (Wn ,τ )
~
ϑ
The second inequality of Lemma 11.19 again goes in the opposite direction to the defining
inequality (126) of a (~, 1 − θ, n) cover.
11.20.4. Completion of Proof. We now sketch the proof of the Theorem (see [A], (2.5.1) (2.5.5)). It is similar to (119) but circumvents the positivity problem by decreasing N to
below the Ehrenfest time.
Proof. (Outline, quoted almost verbatim from [A])
As above, the quantum measure µ~ is shift invariant (see [A], Proposition 1.3.1 (ii).)
We now fix n and let N = κ| log ~|. The crucial point is that this is less than Ehrenfest
time, unlike the choice of N in the first attempt. As above, Wn ⊂ Σn denotes an (~, 1 − θ, n)cover of minimal cardinality of F by n-cylinders. In the following, we use the notation
X
X
(129)
µ~ (Wn ) :=
µ~ (C), µ~ (ΣN (Wn , τ )) :=
µ~ (C).
C∈Wn
We then run through the steps of (119) again;
C∈ΣN (Wn ,τ )
64
STEVE ZELDITCH
|µ~ (Wn )| = | N 1−n
PN −n−1
k=0
µ~ (σ −k Wn )|
¯ ³
´¯
PN −n−1
¯
¯
1
= ¯µ~ N −n k=0 1σ−k Wn ¯
(130)
≤
≤
P
P
C∈ΣN (Wn ,τ )
|µ~ (C)| + τ
C∈ΣN (Wn ,τ )
|µ~ (C)| + τ + O(hκ )
P
C ∈Σ
/ N (Wn ,τ )
|µ~ (C)|
¢
¡
≤ (1 − τ ) 1 − ( ϑκ θ)2 + τ + O(hκ ).
Above, Lemma 11.19 is used to circumvent the positivity problem in (119). We also used
(again) (121) and the main estimate (118) to bound the N -cylinder sets in ΣN (Wn , τ ) with
N = κ| log ~|, but with the time n fixed.
Now let ~ → 0. Since Wn is fixed, the weak convergence µ~ → µ0 implies,
¡
¢
(131)
µ0 (Wn ) ≤ (1 − τ ) 1 − ( ϑκ θ)2 + τ.
Since F ⊂ Wn , the same estimate applies to F . Since it holds for all θ < 1, it follows that
³
κ 2´
κ
(132)
µ0 (F ) ≤ (1 − τ ) 1 − ( ) + τ = 1 − ( )2 (1 − τ ) < 1.
ϑ
ϑ
This completes the proof that supp ν0 6= F . QED
To obtain the measure statement of Theorem 11.5, let
Λ
(133)
Iδ = {x ∈ S ∗ M : hg (µx0 ) ≤ (1 − δ)}.
2
It follows from the Shannon-McMillan theorem that for all α > 0, there exists Iδα ⊂ Iδ with
Λ
µ0 (Iδ \Iδα ) ≤ α such that Iδα can be covered by en( 2 (1−δ+α) n-cylinders for sufficiently large n.
It follows from (132) that
µ
¶
κ
α
2
(134)
µ0 (Iδ ) ≤ (1 − τ (δ − α)) 1 − (
)
+ τ (δ − α).
ϑ(δ − α) +
Let α → 0 and one has,
κ 2
(135)
ν0 (S ∗ M \Iδ ) ≥ (1 − τ (δ))(
).
ϑ(δ)
¤
11.21. F = γ. To help digest the proof, let us run through the proof in the simplest case
where F is a single closed hyperbolic geodesic γ and M is a compact hyperbolic manifold of
constant curvature −1. We would like to show that ν0 6= µγ , or equivalently that ν0 (γ) < 1.
What we need to show is that, if ν0 (F ) = 1, then it requires a lot of cylinder sets to cover F .
Hence we try to get a contradiction from the fact that only a few cylinder sets are needed
to cover γ. In fact, for each n, γ may be covered by one n-cylinder
Wn0 (γ) = {[α0 , . . . , α0 , . . . , αn ]},
namely the one specified by the indices of the cells Pα that γ passes through starting from
some fixed x0 ∈ γ. To obtain a tube around the orbit, we take the union Wn (γ) :=
RECENT DEVELOPMENTS IN MATHEMATICAL QUANTUM CHAOS
65
Sn−1
g −j Wn0 (γ). We could use the two-sided symbols, going symmetrically forward and
backward in time to obtain a transverse ball of radius e−n/2 .
First, let us try a naive heuristic argument. The main estimate (108 ) gives
j=0
(136)
|µ~ (Wn (γ))| ≤ Ch−d/2 e−nd/2 (1 + O(²))n , uniformly for n ≤ K | log ~|,
for any K > 0. It is now tempting to put n = (1 + ²)| log ~| with ² > 0 to obtain
|µ~ (Wn(~,θ) )| ≤ ~θ . But we cannot take the semi-classical limit of this estimate. If we
did so formally (not rigorously), it would seem to imply that ν0 (γ) = 0, which is stronger
than the rigorous result ν0 (γ) < 1 and possibly false in some examples (if weak scarring in
fact occurs).
The problem is that the classical set Wn(~,θ) is a kind of ~-dependent Bowen ball or tube
1
around γ of radius e− 2 n(~,θ) = ~1/2+θ . The uncertainty principle prohibits microlocalization to
1
tubes of radius < h 2 and (as discussed in §5), it is difficult even to work with neighborhoods
1
1
smaller than h 2 −δ or at best h 2 | log ~|. In particular, we do do not have control over the
weak limits of |µ~ (Wn(~,θ) )|, and there is no obvious reason why it should have semi-classical
asymptotics. It is not a positive measure on this length scale, so we do not know if it implies
ν0 (γ) = 0. On the other hand, if we decrease n below the Ehrenfest time, so that the matrix
element has a classical limit, then the bound is trivial.
The strategy of [A] is to use the sub-multiplicative estimate for (~, 1 − θ, N )-covers of γ of
minimal cardinality to decrease the length N below the Ehrenfest time (see Lemma 11.17).
We first consider the cover Wn := Wn (γ), which is a minimal cover of γ by cylinders of
length n. We then form ΣN (Wn , τ ). At least intuitively, Wn corresponds to a Bowen ball of
radius e−n/2 , i.e. a transverse cube of this radius in the stable and unstable directions. The
cylinders of ΣN (Wn , τ ) satisfy σ j (C) ∩ Wn = C for a proportion τ of 0 ≤ j ≤ N . This carves
out a very porous set in the stable/unstable cross section, suggesting why the lower bound
in Lemma 11.19 should be true.
It would be interesting to connect the argument of Theorem 11.5 in this case to the problem
in §5 (see §5.1) of estimating the eigenfunction mass inside a shrinking tubular neighborhood
of γ, i.e. to give an upper bound for the mass (Op~ (χδ1 )ϕµ , ϕµ) ) in the shrinking tube around γ
or (better) the mass (Op~ (χδ2 )ϕµ , ϕµ ) in a shrinking phase space tube. However, the methods
of [A] apply equally to the cat map setting and in that case, there are eigenfunctions whose
mass in shrinking tubes around a hyperbolic fixed point have the same order of magnitude
as in the integrable case up to a factor of 1/2 [FNB]. So unless one can use a property of
∆-eigenfunctions which is not shared by quantum cat map eigenfunctions, one cannot expect
to prove a sharper upper bound for eigenfunction mass in shrinking tubes around hyperbolic
orbits in the Anosov case than one has in the integrable case.
11.22. Some ideas on the main estimate. The main estimate is independently useful.
For instance, it has been used to obtain bounds on decay rates on quantum scattering [NZ].
In this section, we sketch some ideas of the proof. We do not go so far as to explain a key
point, that the estimate is valid for times t ≤ κ| log ~| for any κ.
The goal is to estimate the norm of the operators Pα∗0 U n Pα Op(χ(n) ) in terms of ~ and n.
Since U (n) is unitary it suffices to estimate the norm of the cylinder set operators
Kn := P̂αn U P̂αn −1 · · · U P̂α0 Op(χ).
66
STEVE ZELDITCH
Let Kn (z, w) be the Schwartz kernel of this operator. By the Schur inequality, the operator
norm is bounded above by
Z
sup
|Kn (z, w)|dV (w).
z
M
The inequality only applies when Kn (z, w) ∈ L1 (M ), but by harmless smoothing approximation we may assume this.
We may get an intuitive idea of the exponential decay of the norm of the cylinder set
operators as follows. The wave group (or the semi-classical Schrödinger propagator) U t on a
manifold without conjugate points can be lifted to the universal cover (see (25)) where it has a
global Hadamard parametrix (17). This parametrix shows that the wave
R kernel on the universal cover is closely related to the spherical means operator Lt f (x) = S ∗ M f (g t (x, ω)dµx (ω),
x
where dµx is the Euclidean surface measure on Sx∗ M induced by the metric. Both operators
have the same wave front relation, and indeed the wave front set of U (t, ·, y) is precisely
the set of (co-)normal directions to the distance sphere St (y) centered at y. The operator
relating U t and Lt is a dth order (pseudo-differential) operator in (t, x), and in the Fourier
transformed picture in [A] it gives rise to the factor h−d/2 . As time evolves, the surface volume of the distance sphere in the universal cover grows exponentially. Since U t is unitary,
its amplitude must decay exponentially. In fact this is evident from (17) since Θ(x, y) grows
exponentially in the distance from x to y, so Θ−1/2 (x, y) decays exponentially.
The cylinder set operators Cˆ~ = P̂αn (n) · · · P̂α1 (1)P̂α0 (0) are given by a sequence of alternating applications of the propagator U and the cutting off operators Pαj (which are
multiplications). At least heuristically, in estimating the Schur integral for Kn (x, y), one is
first cutting off to a piece of the distance of the distance sphere S1 (y) of radius one centered
at y. Then one applies U and this piece expands to S2 (y). Then one cuts it off with Pα2 . As
this picture evolves, one ends up after n iterates with a piece of the distance sphere Sn (y)
of radius n centered at y which is has the same size as the initial piece. But the spherical
means and wave group are normalized by dividing by the volume, so the contribution to the
Schur integral is exponentially decaying at the rate e−n .
11.23. Clarifications to [A].
(1) The θ on p. 447 is a new parameter < 1 which is implicitly defined above (2.0.1).
(2) The θ in (2.3.2) is also implicitly defined by this inequality. Wn1 is not assumed to
be a θ cover.
(3) In (2.5.1) – (2.5.5), the Σ in σ −k Σ(Wn1 ) should be removed.
mentioned above, the notation µh (ΣN (Wn , τ )) is not quite defined but means
(4) As S
µh ( C∈ΣN (Wn ,τ )) C).
12. Applications to nodal hypersurfaces of eigenfunctions
This survey has been devoted to the asymptotics of matrix elements hAϕj , ϕj i. In the
final section, we address the question, “what applications does the study of these matrix
elements have?”
In automorphic forms, the matrix elements are related to L-functions, Rankin-Selberg zeta
functions, etc. So there is ample motivation for their study in the arithmetic case. Such
applications lie outside the scope of this survey, so we refer the reader to Sarnak’s expository
RECENT DEVELOPMENTS IN MATHEMATICAL QUANTUM CHAOS
67
articles [Sar2, Sar3] for background on and references to the large literature in arithmetic
quantum chaos.
Here we are concerned with applications to the classical and geometric analysis of eigenfunctions on general Riemannian manifolds. We present an application of quantum ergodicity
to the study of nodal (zero) sets of eigenfunctions on real analytic (M, g) with ergodic geodesic flow. It is not implausible that one could use related methods to count critical points
of eigenfunctions in the ergodic case.
The nodal hypersurface of ϕj is the zero set Nj = {x ∈ M : ϕj (x) = 0}. We are interested
in its distribution as λj → ∞. We study the distribution through the integrals
Z
f dS
Nj
of f over the nodal line with respect to the natural surface measure. If f = 1U is the
characteristic (indicator) function of an open set U ⊂ M , then
Z
(137)
1U ds = Hn−1 (U ∩ Nj )
Nj
is (n−1) dimensional Hausdorff measure of the part of Nj which intersects U . When U = M
the integral (137) gives the total surface volume of the nodal set. The principal result on
volumes is due to Donnelly-Fefferman [DF], proving a conjecture of S. T. Yau [Y1]:
Theorem 12.1. [DF] (see also [Lin]) Let (M, g) be a compact real analytic Riemannian
manifold, with or without boundary. Then there exist c1 , C2 depending only on (M, g) such
that
(138)
c1 λ ≤ Hm−1 (Zϕλ ) ≤ C2 λ,
(∆ϕλ = λ2 ϕλ ; c1 , C2 > 0).
The question is whether one can obtain asymptotics results by imposing a global dynamical
condition on the geodesic flow.
The hypothesis of real analyticity is crucial: The analysis is based on the analytic continuation of eigenfunctions to the complexification MC of M . The articles [DF, Lin] also used
analytic continuation to study volumes of nodal hypersurfaces. It is the role of ergodicity in
the complex setting that gives strong equidistribution results.
12.1. Complexification of (M, g) and Grauert tubes. By a theorem of Bruhat and
Whitney, a real analytic manifold M always possesses a unique complexification MC , an open
complex manifold in which M embeds ι : M → MC as a totally real submanifold. Examples
are real algebraic subvarieties of Rn which complexify to complex algebraic subvarieties
of Cn . Other simple examples include the complexification of a quotient such as M =
Rm /Zm to MC = Cm /Zm . The complexification of hyperbolic space can be defined using the
hyperboloid model Hn = {x21 + · · · x2n − x2n+1 = −1, xn > 0}. Then, HnC = {(z1 , . . . , zn+1 ) ∈
2
= −1}.
Cn+1 : z12 + · · · zn2 − zn+1
The Riemannian metric determines a special kind of plurisubharmonic exhaustion function
√
√
on MC [GS1, GS2, LemS1], namely the Grauert tube radius ρ = ρg on MC , defined as
the unique solution of the Monge-Ampère equation
√
¯ = g.
(∂ ∂¯ ρ)m = δMR ,dVg , ι∗ (i∂ ∂ρ)
68
STEVE ZELDITCH
Here, δMR ,dVg is the delta-function on the real M with
p respect to the volume form dVg , i.e.
R
√
f → M f dVg . It is shown in [GS1] that ρ(ζ) = i r2 (ζ, ζ̄) where r2 (x, y) is the squared
distance function in a neighborhood of the diagonal in M × M .
The Grauert tubes Mτ is defined by
√
Mτ = {ζ ∈ MC : ρ(ζ) ≤ τ }.
√
The maximal value of τ0 for which ρ is well defined is known as the Grauert tube radius.
For τ ≤ τ0 , Mτ is a strictly pseudo-convex domain in MC . Using the complexified exponential
map (x, ξ) → expx iξ one may identify Mτ with the co-ball bundle Bτ∗ M to Mτ . Under this
√
√
identification, ρ corresponds to |ξ|g . The one-complex dimensional null foliation of ∂ ∂¯ ρ,
known as the ‘Monge-Ampère’ or Riemann foliation, are the complex curves t + iτ → τ γ̇(t),
where γ is a geodesic. The geometric Grauert tube radius is the maximal radius for which
the exponential map has a holomorphic extension and defines a diffeomorphism.
12.2. Analytic Continuation of eigenfunctions. The eigenfunctions are real analytic
and therefore possess analytic continuations to some Grauert tube M² independent of the
eigenvalue. To study analytic continuation of eigenfunctions, it is useful to relate
them to
√
−τ ∆
analytic continuation of the wave group at imaginary time, U (iτ, x, y) = e
(x, y). This
is the Poisson operator of (M, g). We denote its analytic continuation in the x variable to
M² by UC (iτ, ζ, y). In terms of the eigenfunction expansion, one has
(139)
U (iτ, ζ, y) =
∞
X
e−τ λj ϕCj (ζ)ϕj (y), (ζ, y) ∈ M² × M.
j=0
To the author’s knowledge, the largest ² has not been determined at this time. In particular, it is not known if the radius of analytic continuation of the wave kernel and of the
eigenfunctions is the same as the geometrically defined Grauert tube radius.
Since
(140)
UC (iτ )ϕλ = e−τ λ ϕCλ ,
the analytic continuability of the Poisson operator to Mτ implies that every eigenfunction
analytically continues to the same Grauert tube. It follows that the analytic continuation
operator to Mτ is given by
(141)
AC (τ ) = UC (iτ ) ◦ eτ
√
∆
.
√
Thus, a function f ∈ C ∞
(M ) has a holomorphic
extension to the closed tube ρ(ζ) ≤ τ if
√
√
√
and only if f ∈ Dom(eτ ∆ ), where eτ ∆ is the backwards ‘heat operator’ generated by ∆
(rather than ∆).
Let us consider examples of holomorphic continuations of eigenfunctions:
• On the flat torus Rm /Zm , the real eigenfunctions are coshk, xi, sinhk, xi with k ∈
2πZm . The complexified torus is Cm /Zm and the complexified eigenfunctions are
coshk, ζi, sinhk, ζi with ζ = x + iξ.
• On the unit sphere S m , eigenfunctions are restrictions of homogeneous harmonic
functions on Rm+1 . The latter extend holomorphically to holomorphic harmonic
polynomials on Cm+1 and restrict to holomorphic function on SCm .
RECENT DEVELOPMENTS IN MATHEMATICAL QUANTUM CHAOS
69
• On Hm , one may use the hyperbolic plane waves e(iλ+1)hz,bi , where hz, bi is the (signed)
hyperbolic distance of the horocycle passing through z and b to 0. They may be
holomorphically extended to the maximal tube of radius π/4. See e.g. [BR].
• On compact hyperbolic quotients Hm /Γ, eigenfunctions can be then represented by
Helgason’s generalized Poisson integral formula,
Z
ϕλ (z) =
e(iλ+1)hz,bi dTλ (b).
B
To analytically continue ϕλ it suffices to analytically continue hz, bi. Writing the latter
as hζ, bi, we have:
Z
C
(142)
ϕλ (ζ) =
e(iλ+1)hζ,bi dTλ (b).
B
• In the case of 1D Schrödinger operators, several authors have studied analytic continuations of eigenfunctions and their complex zeros, see e.g. [Hez, EGS].
12.3. Maximal plurisubharmonic functions and growth of ϕCλ . In the case of domains
Ω ⊂ Cm , the maximal PSH (pluri-subharmonic) function (or pluri-complex Green’s function)
relative to a subset E ⊂ Ω is defined by
VE (ζ) = sup{u(z) : u ∈ P SH(Ω), u|E ≤ 0, u|∂Ω ≤ 1}.
An alternative definition due to Siciak takes the supremum only with respect to polynomials
p. We denote by P N the space of all complex analytic polynomials of degree N and put
N
PK
= {p ∈ P N : ||p||K ≤ 1, ||p||Ω ≤ e}. Then define
1
log ΦN
log |pN (ζ)| : p ∈ PEN }, log ΦE = lim sup log ΦN
E (ζ) = sup{
E.
N
N →∞
Here, ||f ||K = supz∈K |f (z)|. Siciak proved that log ΦE = VE .
On a real analytic Riemannian manifold, the natural analogue of P N is the space
X
aj ϕλj , a1 , . . . , aN (λ) ∈ R}
Hλ = {p =
j:λj ≤λ
spanned by eigenfunctions with frequencies ≤ λ. Rather than using the sup norm, it is
convenient to work with L2 based norms than sup norms, and so we define
λ
HM
= {p =
N (λ)
X
aj ϕλj , ||p||L2 (M ) =
X
|aj |2 = 1}.
j=1
j:λj ≤λ
We define the λ-Siciak approximate extremal function by
ΦλM (z) = sup{|ψ(z)|1/λ : ψ ∈ Hλ ; kψkM 6 1},
and the extremal function by
ΦM (z) = sup ΦλM (z).
λ
It is not hard to show that
(143)
1
√
log Π[0,λ (ζ, ζ̄) = ρ.
λ→∞ λ
ΦM (z) = lim
70
STEVE ZELDITCH
12.4. Nodal hypersurfaces in the case of ergodic geodesic flow. The complex nodal
hypersurface of an eigenfunction is defined by
ZϕCλ = {ζ ∈ B²∗0 M : ϕCλ (ζ) = 0}.
(144)
There exists a natural current of integration over the nodal hypersurface in any ball bundle
B²∗ M with ² < ²0 , given by
Z
Z
i
C 2
¯
(145)
h[ZϕCλ ], ϕi =
∂ ∂ log |ϕλ | ∧ ϕ =
ϕ, ϕ ∈ D(m−1,m−1) (B²∗ M ).
2π B²∗ M
Z C
ϕ
λ
In the second equality we used the Poincaré-Lelong formula. The notation D(m−1,m−1) (B²∗ M )
stands for smooth test (m − 1, m − 1)-forms with support in B²∗ M.
The nodal hypersurface ZϕCλ also carries a natural volume form |ZϕCλ | as a complex hypersurface in a Kähler manifold. By Wirtinger’s formula, it equals the restriction of
ZϕCλ .
ωgm−1
(m−1)!
to
Theorem 12.2. Let (M, g) be real analytic, and let {ϕjk } denote a quantum ergodic sequence
of eigenfunctions of its Laplacian ∆. Let (B²∗0 M, J) be the maximal Grauert tube around M
with complex structure Jg adapted to g. Let ² < ²0 . Then:
i √
1
0
[ZϕCj ] → ∂ ∂¯ ρ weakly in D (1,1) (B²∗ M ),
k
λjk
π
in the sense that, for any continuous test form ψ ∈ D(m−1,m−1) (B²∗ M ), we have
Z
Z
i
1
√
ψ→
ψ ∧ ∂ ∂¯ ρ.
λjk Z C
π B²∗ M
ϕ
jk
Equivalently, for any ϕ ∈ C(B²∗ M ),
Z
Z
ωgm−1
ωgm−1
1
i
√
¯
ϕ
→
ϕ∂ ∂ ρ ∧
.
λjk Z C (m − 1)!
π B²∗ M
(m − 1)!
ϕ
jk
Corollary 12.3. Let (M, g) be a real analytic with ergodic geodesic flow. Let {ϕjk } denote
a full density ergodic sequence. Then for all ² < ²0 ,
1
i √
0
[ZϕCj ] → ∂ ∂¯ ρ, weakly in D (1,1) (B²∗ M ).
k
λjk
π
¯ on MC adapted to
Remark: The limit form is singular with respect to the Kählerform i∂ ∂ρ
the metric. Its highest exterior power is the delta function on the zero section (i.e. the real
domain M ).
In outline the steps in the proof are;
(1) By the Poincaré-Lelong formula, [ZϕCλ ] = i∂ ∂¯ log |ϕCλ |. This reduces the theorem to
determining the limit of λ1 log |ϕCλ |.
(2) |ϕCλ |2 , when properly L2 normalized on each ∂Mτ is a quantum ergodic sequence on
∂Mτ .
RECENT DEVELOPMENTS IN MATHEMATICAL QUANTUM CHAOS
71
(3) The ergodic property of complexified eigenfunctions implies that the L2 norm of |ϕCλ |2
√
on ∂Mτ is asymtotically ρ. Thus, normalized log moduli of ergodic eigenfunctions
are asympotically maximal PSH functions. They are λ-Siciak extremal functions.
|ϕC (ζ)|2
(4) λ1 log ||ϕC ||λ2
→ 0 for an ergodic sequence of eigenfunctions. Hence,
λ L2 (∂Mτ
√
(5) λi ∂ ∂¯ log |ϕCλ | → i∂ ∂¯ ρ,
concluding the proof.
Remark: We expect a similar result for nodal lines of Dirichlet or Neumann eigenfunctions
of a piecewise smooth domain Ω ⊂ R2 with ergodic billiards. We also expect similar results
for nodal lines of such domains which touch the boundary (like the angular ‘spokes’ of
eigenfunctions of the disc). In [TZ1], we proved that the number of such nodal lines is
bounded above by CΩ λ. This is rather like the Donnelly-Fefferman upper bound. However
the example of the disc shows that there is no lower bound above zero. That is, there
are sequences of eigenfunctions (with low angular momentum) with a uniformly bounded
number of nodal lines touching the boundary. We believe that ergodicity is a mechanism
producing lower bounds on numbers of zeros and critical points. It causes eigenfunctions to
oscillate uniformly in all directions and hence to produce a lot of zeros and critical points.
We expect this to be one of the future roles for ergodicity of eigenfunctions.
Remark: Precursors to Theorem 12.2 giving limit distributions of zero sets for holomorphic
eigen-sections of Kählerquantizations of ergodic symplectic maps on Kählermanifolds can be
found in [NV2, SZ]. In that setting, the zeros become uniformly distributed with respect to
the Kähler form. In the Riemannian setting, they become uniformly distributed with respect
¯ on MC .
to a singular (1, 1) current relative to the Kähler form ∂ ∂ρ
References
[A]
[A2]
[ANK]
[AN]
[AN2]
[AZ]
[AS]
[B.B]
[BSS]
[Bar]
[Be]
N. Anantharaman, Entropy and the localization of eigenfunctions, Ann. of Math. (2) 168 (2008),
no. 2, 435–475.
N. Anantharaman, Eigenfunctions of the Laplacian on negatively curved manifolds: a semiclassical approach, Lecture Notes (online at http://www.math.polytechnique.fr/ nalini/).
N. Anantharaman, H. Koch, and S. Nonnenmacher, Entropy of eigenfunctions, arXiv:0704.1564.
N. Anantharaman and S. Nonnenmacher, Half-delocalization of eigenfunctions for the Laplacian
on an Anosov manifold, Annales Inst. Fourier 57, number 6 (2007), 2465-2523 (arXiv:mathph/0610019).
N. Anantharaman and S. Nonnenmacher, Vibrations chaotiques et grosses balafres. Gaz. Math.
No. 119 (2009), 16–32. English Version (different): Chaotic vibrations and strong scars.
N. Anantharaman and S. Zelditch, Patterson-Sullivan distributions and quantum ergodicity. Ann.
Henri Poincar 8 (2007), no. 2, 361–426.
R. Aurich and F. Steiner, Statistical properties of highly excited quantum eigenstates of a strongly
chaotic system. Phys. D 64 (1993), no. 1-3, 185–214.
V.M.Babic, V.S. Buldyrev: Short-Wavelength Diffraction Theory, Springer Series on Wave Phenomena 4, Springer-Verlag, New York (1991).
A. Backer, R. Schubert, and P. Stifter, On the number of bouncing ball modes in billiards. J.
Phys. A 30 (1997), no. 19, 6783–6795.
A. Barnett, Asymptotic rate of quantum ergodicity in chaotic Euclidean billiards, Comm. Pure
Appl. Math. 59 (2006), no. 10, 1457–1488.
P. Bérard, On the wave equation without conjugate points, Math. Zeit. 155 (1977), 249–276.
72
[BR]
[Ber]
[BL]
[BouR]
[Bu]
[BZ]
[Chr]
[CV]
[CV2]
[CV3]
[CVP]
[CR]
[DSj]
[DF]
[Dui]
[D.G]
[EGS]
[EZ]
[FN]
[FNB]
[FP]
[GL]
[GSj]
[GS1]
[GS2]
[GSt]
STEVE ZELDITCH
J. Bernstein and A. Reznikov, Analytic continuation of representations and estimates of automorphic forms. Ann. of Math. (2) 150 (1999), no. 1, 329–352.
M. Berry, Regular and irregular semiclassical wavefunctions, J. Phys. A 10 (1977), 2083–2091.
J. Bourgain and E. Lindenstrauss, Entropy of quantum limits. Comm. Math. Phys. 233 (2003),
no. 1, 153–171.
A. Bouzouina and D. Robert, Uniform semiclassical estimates for the propagation of quantum
observables. Duke Math. J. 111 (2002), no. 2, 223–252.
N. Burq, Quantum ergodicity of boundary values of eigenfunctions: A control theory approach,
Canad. Math. Bull. 48 (2005), no. 1, 3–15 (math.AP/0301349).
N. Burq and M. Zworski, Geometric control in the presence of a black box. J. Amer. Math. Soc.
17 (2004), no. 2, 443–471.
H. Christianson, Semiclassical non-concentration near hyperbolic orbits. J. Funct. Anal. 246
(2007), no. 2, 145–195.
Y.Colin de Verdière, Ergodicité et fonctions propres du Laplacien, Comm.Math.Phys. 102 (1985),
497-502.
———,Quasi-modes sur les varietes Riemanniennes compactes, Invent.Math. 43 (1977), 15-52.
, ———, Semi-classical measures and entropy [after Nalini Anantharaman and Stéphane Nonnenmacher]. Séminaire Bourbaki. Vol. 2006/2007. Astérisque No. 317 (2008), Exp. No. 978, ix,
393–414.
Y. Colin de Verdière and B. Parisse, Équilibre instable en régime semi-classique. I. Concentration
microlocale, Comm. in PDE 19 (1994), no. 9-10, 1535–1563.
M. Combescure and D. Robert, Semiclassical spreading of quantum wave packets and applications
near unstable fixed points of the classical flow, Asymptot. Anal. 14 (1997), 377–404.
M. Dimassi and J. Sjöstrand, Spectral asymptotics in the semi-classical limit. London Mathematical Society Lecture Note Series, 268. Cambridge University Press, Cambridge, 1999.
H. Donnelly and C. Fefferman, Nodal sets of eigenfunctions on Riemannian manifolds, Invent.
Math. 93 (1988), 161-183.
J. J. Duistermaat, Oscillatory integrals, Lagrange immersions and unfolding of singularities.
Comm. Pure Appl. Math. 27 (1974), 207–281.
J.J.Duistermaat and V.Guillemin, The spectrum of positive elliptic operators and periodic bicharacteristics, Inv.Math. 24 (1975), 39-80.
A. Eremenko, A. Gabrielov, and B. Shapiro, High energy eigenfunctions of one-dimensional
Schrödinger operators with polynomial potentials. Comput. Methods Funct. Theory 8 (2008),
no. 1-2, 513–529.
L. Evans and M. Zworski, Introduction to semiclassical analysis, UC Berkeley lecture notes (2006).
http://math.berkeley.edu/ zworski.
F. Faure and S. Nonnenmacher, On the maximal scarring for quantum cat map eigenstates.
Comm. Math. Phys. 245 (2004), no. 1, 201–214.
F. Faure, S. Nonnenmacher, and S. De Bièvre, Scarred eigenstates for quantum cat maps of
minimal periods. Comm. Math. Phys. 239 (2003), no. 3, 449–492.
M. Feingold and A. Peres, Distribution of matrix elements of chaotic systems. Phys. Rev. A (3)
34 (1986), no. 1, 591–595.
P.Gérard and E.Leichtnam, Ergodic properties of eigenfunctions for the Dirichlet problem, Duke
Math J. 71 (1993), 559-607.
A. Grigis and J. Sjöstrand, Microlocal analysis for differential operators, London Math. Soc.
Lecture Notes 196 (1994).
V. Guillemin and M. Stenzel, Grauert tubes and the homogeneous Monge-Ampre equation. J.
Differential Geom. 34 (1991), no. 2, 561–570.
———–, Grauert tubes and the homogeneous Monge-Ampre equation. II. J. Differential Geom.
35 (1992), no. 3, 627–641.
V. Guillemin and S. Sternberg, Geometric asymptotics. Mathematical Surveys, No. 14. American
Mathematical Society, Providence, R.I., 1977.
RECENT DEVELOPMENTS IN MATHEMATICAL QUANTUM CHAOS
[Gut]
[HanL]
[Has]
[HZ]
[HR]
[Hel]
[He]
[He2]
[HO]
[Hez]
[Hol]
[HS]
[HoI-IV]
[J]
[Kl]
[KR]
[L]
[LemS1]
[Lin]
[LIND]
[LIND2]
[LS]
[LS2]
[M]
[MOZ]
[MS]
73
B. Gutkin, Note on converse quantum ergodicity. Proc. Amer. Math. Soc. 137 (2009), no. 8,
2795–2800.
Q. Han and F.H. Lin Nodal sets of solutions of elliptic differential equations, book in preparation
(2007).
A. Hassell, Ergodic billiards that are not quantum unique ergodic, Appendix by A. Hassell and
L. Hillairet, to appear in Ann. Math. (arXiv:0807.0666).
A. Hassell and S. Zelditch, Quantum ergodicity of boundary values of eigenfunctions. Comm.
Math. Phys. 248 (2004), no. 1, 119–168.
D. A. Hejhal and B. N. Rackner, On the topography of Maass waveforms for PSL(2, Z). Experiment. Math. 1 (1992), no. 4, 275–305.
S. Helgason, Topics in harmonic analysis on homogeneous spaces. Progress in Mathematics, 13.
Birkhuser, Boston, Mass., 1981.
E.J. Heller, Bound-state eigenfunctions of classically chaotic Hamiltonian systems: scars of periodic orbits. Phys. Rev. Lett. 53 (1984), no. 16, 1515–1518.
E.J. Heller, Wavepacket dynamics and quantum chaology. Chaos et physique quantique (Les
Houches, 1989), 547–664, North-Holland, Amsterdam, 1991.
E. J. Heller and P. W. O’Connor, Quantum localization for a strongly classical chaotic system,
Phys. Rev. Lett. 61 (20) (1988), 2288-2291.
H. Hezari, Complex zeros of eigenfunctions of 1D Schrdinger operators. Int. Math. Res. Not.
IMRN 2008, no. 3, Art. ID rnm148, 23 pp.
R. Holowinsky, Sieving for mass equidistribution, to appear in Annals of Math. (arXiv:0809.1640).
R. Holowinsky and K. Soundararajan, Mass equidistribution for Hecke eigenforms, to appear in
Annals of Math. (arXiv:0809.1636)
L. Hörmander, Theory of Linear Partial Differential Operators I-IV, Springer-Verlag, New York
(1985).
O. Jenkinson, Every ergodic measure is uniquely maximizing, Disc. Cont. Dyn. Syst. 16 (2006).
W. Klingenberg, Lectures on Closed Geodesics, Grundlehren der. math. W. 230, Springer-Verlag
(1978).
P. Kurlberg and Z. Rudnick, On the distribution of matrix elements for the quantum cat map.
Ann. of Math. (2) 161 (2005), no. 1, 489–507.
V. F. Lazutkin, KAM theory and semiclassical approximations to eigenfunctions. With an addendum by A. I. Shnirelman. Ergebnisse der Mathematik und ihrer Grenzgebiete (3), 24. SpringerVerlag, Berlin, 1993.
L. Lempert and R. Szöke, Global solutions of the homogeneous complex Monge-Ampre equation
and complex structures on the tangent bundle of Riemannian manifolds. Math. Ann. 290 (1991),
no. 4, 689–712.
F.H. Lin, Nodal sets of solutions of elliptic and parabolic equations. Comm. Pure Appl. Math. 44
(1991), no. 3, 287–308.
E. Lindenstrauss, Invariant measures and arithmetic quantum ergodicity, Annals of Math. (2) 163
(2006), no. 1, 165–219.
E. Lindenstrauss, Adelic dynamics and arithmetic quantum unique ergodicity. Current developments in mathematics, 2004, 111–139, Int. Press, Somerville, MA, 2006.
W.Luo and P.Sarnak, Quantum ergodicity of eigenfunctions on PSL2 (Z)\H2 , IHES Publ. 81
(1995), 207-237.
———–, Quantum variance for Hecke eigenforms, Annales Scient. de l’École Norm. Sup. 37 (2004),
p. 769-799.
J. Marklof, Arithmetic quantum chaos, Encyclopedia of Mathematical Physics article 449.
J. Marklof, S. O’Keefe, Weyl’s law and quantum ergodicity for maps with divided phase space.
With an appendix ”Converse quantum ergodicity” by Steve Zelditch. Nonlinearity 18 (2005), no.
1, 277–304.
S. D. Miller and W. Schmid, The highly oscillatory behavior of automorphic distributions for
SL(2). Lett. Math. Phys. 69 (2004), 265–286.
74
[NJT]
STEVE ZELDITCH
N. Nadirashvili, D. Jakobson, and J.A. Toth, Geometric properties of eigenfunctions. (Russian)
Uspekhi Mat. Nauk 56 (2001), no. 6(342), 67–88; translation in Russian Math. Surveys 56 (2001),
no. 6, 1085–1105
[NV]
S. Nonnenmacher and A. Voros, Eigenstate structures around a hyperbolic point. J. Phys. A 30
(1997), no. 1, 295–315.
[NV2]
—————————, Chaotic eigenfunctions in phase space. J. Statist. Phys. 92 (1998), no. 3-4,
431–518.
N. Nonnenmacher and M. Zworski, Quantum decay rates in chaotic scattering, to appear in Acta.
[NZ]
Math. ( arXiv:0706.3242).
[O]
J.P. Otal, Sur les fonctions propres du laplacien du disque hyperbolique. (French. English, French
summary) [About eigenfunctions of the Laplacian on the hyperbolic disc] C. R. Acad. Sci. Paris
Sér. I Math. 327 (1998), no. 2, 161–166.
[Ra]
J. Ralston, Gaussian beams and the propagation of singularities. Studies in partial differential
equations, 206–248, MAA Stud. Math., 23, Math. Assoc. America, Washington, DC, 1982.
[Ra2]
———–, Approximate eigenfunctions of the Laplacian. J. Differential Geometry 12 (1977), no. 1,
87–100.
G. Riviere, Entropy of semiclassical measures in dimension 2 ( arXiv:0809.0230).
[Riv]
[RS]
Z. Rudnick and P. Sarnak, The behaviour of eigenstates of arithmetic hyperbolic manifolds.
Comm. Math. Phys. 161 (1994), no. 1, 195–213.
[Ru]
D. Ruelle, Statistical mechanics: Rigorous results. W. A. Benjamin, Inc., New York-Amsterdam
1969
[Sar]
P. Sarnak, Integrals of products of eigenfunctions. Internat. Math. Res. Notices no. 6(1997), 251261.
[Sar2]
———–, Arithmetic quantum chaos. The Schur lectures (1992) (Tel Aviv), 183–236, Israel Math.
Conf. Proc., 8, Bar-Ilan Univ., Ramat Gan, 1995.
1926.
[Sar3]
————, Recent progress on QUE (to appear).
[Sch]
W. Schmid, Automorphic distributions for SL(2, R). (English summary) Confrence Mosh Flato
1999, Vol. I (Dijon), 345–387, Math. Phys. Stud., 21, Kluwer Acad. Publ., Dordrecht, 2000.
[Schu]
R. Schubert, On the rate of quantum ergodicity for quantised maps. Ann. Henri Poincaré 9 (2008),
no. 8, 1455–1477
[Schu2]
. ———-, Upper bounds on the rate of quantum ergodicity. Ann. Henri Poincaré 7 (2006), no. 6,
1085–1098.
[Schu3]
———–, Semiclassical behaviour of expectation values in time evolved Lagrangian states for large
times. Comm. Math. Phys. 256 (2005), no. 1, 239–254.
[Sh.1]
A.I.Shnirelman, Ergodic properties of eigenfunctions, Usp.Math.Nauk. 29/6 (1974), 181-182.
[Sh.2]
———–, On the asymptotic properties of eigenfunctions in the region of chaotic motion, addendum to V.F.Lazutkin, KAM theory and semiclassical approximations to eigenfunctions, Springer
(1993).
[SZ]
B. Shiffman and S. Zelditch, Distribution of zeros of random and quantum chaotic sections of
positive line bundles. (English summary) Comm. Math. Phys. 200 (1999), no. 3, 661–683.
[Sound1] K. Soundararajan, Quantum unique ergodicity for SL2 (Z)H, to appear in Annals of Math.
(arXiv:0901.4060)
T. Sunada, Quantum ergodicity. Progress in inverse spectral geometry, 175–196, Trends Math.,
[Su]
Birkhuser, Basel, 1997.
J. A. Toth and S. Zelditch, Counting nodal lines which touch the boundary of an analytic domain.
[TZ1]
J. Differential Geom. 81 (2009), no. 3, 649–686
[TZ2]
————————-, Lp norms of eigenfunctions in the completely integrable case. Ann. Henri
Poincaré 4 (2003), no. 2, 343–368.
————————, Quantum ergodic restriction theorems, I: interior hypersurfaces in analytic
[TZ3]
domains.
RECENT DEVELOPMENTS IN MATHEMATICAL QUANTUM CHAOS
[Wat]
[W]
[W2]
[W3]
[Y1]
[Z1]
[Z2]
[Z3]
[Z4]
[Z5]
[Z6]
[Z7]
[Z8]
[Z9]
[Z10]
[ZZw]
[Zh]
75
T. Watson, Central Value of Rankin Triple L-function for Unramified Maass Cusp Forms, to
appear in Annals of Math.
S. A. Wolpert, Semiclassical limits for the hyperbolic plane. Duke Math. J. 108 (2001), no. 3,
449–509.
———–, Asymptotic relations among Fourier coefficients of automorphic eigenfunctions. Trans.
Amer. Math. Soc. 356 (2004), no. 2, 427–456.
———–, The modulus of continuity for Γ0 (m)\H semi-classical limits, Comm. Math. Phys. 216
(2001), 313–323.
S.T. Yau, Survey on partial differential equations in differential geometry. Seminar on Differential
Geometry, pp. 3–71, Ann. of Math. Stud., 102, Princeton Univ. Press, Princeton, N.J., 1982.
S. Zelditch, Local and global analysis of eigenfunctions, in: Handbook of Geometric Analysis, No.
1 L. Ji, P. Li, R. Schoen and L. Simon (eds.), Somerville, MA: International Press; Beijing: Higher
Education Press. Advanced Lectures in Mathematics (ALM) 7, 545-658 (2008).
———–, Quantum ergodicity and mixing, Encyclopedia of Mathematical Physics Vol. 4, Ed. J.P.
Franoise, G. Naber, S.T. Tsou (2007), 183-196.
———, Uniform distribution of eigenfunctions on compact hyperbolic surfaces. Duke Math. J. 55
(1987), no. 4, 919–941
———–, On the rate of quantum ergodicity. I. Upper bounds. Comm. Math. Phys. 160 (1994),
no. 1, 81–92
———, Complex zeros of real ergodic eigenfunctions. Invent. Math. 167 (2007), no. 2, 419–443.
———, Quantum ergodicity of C ∗ dynamical systems. Comm. Math. Phys. 177 (1996), no. 2,
507–528.
———, Note on quantum unique ergodicity, Proc. Amer. Math. Soc. 132 (2004), no. 6, 1869–1872.
———, Mean Lindelöf hypothesis and equidistribution of cusp forms and Eisenstein series. J.
Funct. Anal. 97 (1991), no. 1, 1–49.
———-, Quantum transition amplitudes for ergodic and for completely integrable systems. J.
Funct. Anal. 94 (1990), no. 2, 415–436.
———, On a “quantum chaos” theorem of R. Schrader and M. Taylor. J. Funct. Anal. 109 (1992),
no. 1, 1–21.
S.Zelditch and M.Zworski, Ergodicity of eigenfunctions for ergodic billiards, Comm.Math. Phys.
175 (1996), 673-682.
P. Zhao, Quantum Variance of Maass-Hecke Cusp Forms, to appear in Comm. Math. Phys.
Department of Mathematics, Johns Hopkins University, Baltimore, MD 21218, USA
E-mail address: [email protected]